2 * sd.c Copyright (C) 1992 Drew Eckhardt
3 * Copyright (C) 1993, 1994, 1995, 1999 Eric Youngdale
5 * Linux scsi disk driver
6 * Initial versions: Drew Eckhardt
7 * Subsequent revisions: Eric Youngdale
8 * Modification history:
9 * - Drew Eckhardt <drew@colorado.edu> original
10 * - Eric Youngdale <eric@andante.org> add scatter-gather, multiple
11 * outstanding request, and other enhancements.
12 * Support loadable low-level scsi drivers.
13 * - Jirka Hanika <geo@ff.cuni.cz> support more scsi disks using
14 * eight major numbers.
15 * - Richard Gooch <rgooch@atnf.csiro.au> support devfs.
16 * - Torben Mathiasen <tmm@image.dk> Resource allocation fixes in
17 * sd_init and cleanups.
18 * - Alex Davis <letmein@erols.com> Fix problem where partition info
19 * not being read in sd_open. Fix problem where removable media
20 * could be ejected after sd_open.
21 * - Douglas Gilbert <dgilbert@interlog.com> cleanup for lk 2.5.x
22 * - Badari Pulavarty <pbadari@us.ibm.com>, Matthew Wilcox
23 * <willy@debian.org>, Kurt Garloff <garloff@suse.de>:
24 * Support 32k/1M disks.
26 * Logging policy (needs CONFIG_SCSI_LOGGING defined):
27 * - setting up transfer: SCSI_LOG_HLQUEUE levels 1 and 2
28 * - end of transfer (bh + scsi_lib): SCSI_LOG_HLCOMPLETE level 1
29 * - entering sd_ioctl: SCSI_LOG_IOCTL level 1
30 * - entering other commands: SCSI_LOG_HLQUEUE level 3
31 * Note: when the logging level is set by the user, it must be greater
32 * than the level indicated above to trigger output.
35 #include <linux/module.h>
37 #include <linux/kernel.h>
39 #include <linux/bio.h>
40 #include <linux/genhd.h>
41 #include <linux/hdreg.h>
42 #include <linux/errno.h>
43 #include <linux/idr.h>
44 #include <linux/interrupt.h>
45 #include <linux/init.h>
46 #include <linux/blkdev.h>
47 #include <linux/blkpg.h>
48 #include <linux/delay.h>
49 #include <linux/mutex.h>
50 #include <linux/string_helpers.h>
51 #include <linux/async.h>
52 #include <linux/slab.h>
53 #include <linux/sed-opal.h>
54 #include <linux/pm_runtime.h>
56 #include <linux/t10-pi.h>
57 #include <linux/uaccess.h>
58 #include <asm/unaligned.h>
60 #include <scsi/scsi.h>
61 #include <scsi/scsi_cmnd.h>
62 #include <scsi/scsi_dbg.h>
63 #include <scsi/scsi_device.h>
64 #include <scsi/scsi_driver.h>
65 #include <scsi/scsi_eh.h>
66 #include <scsi/scsi_host.h>
67 #include <scsi/scsi_ioctl.h>
68 #include <scsi/scsicam.h>
71 #include "scsi_priv.h"
72 #include "scsi_logging.h"
74 MODULE_AUTHOR("Eric Youngdale");
75 MODULE_DESCRIPTION("SCSI disk (sd) driver");
76 MODULE_LICENSE("GPL");
78 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK0_MAJOR);
79 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK1_MAJOR);
80 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK2_MAJOR);
81 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK3_MAJOR);
82 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK4_MAJOR);
83 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK5_MAJOR);
84 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK6_MAJOR);
85 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK7_MAJOR);
86 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK8_MAJOR);
87 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK9_MAJOR);
88 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK10_MAJOR);
89 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK11_MAJOR);
90 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK12_MAJOR);
91 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK13_MAJOR);
92 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK14_MAJOR);
93 MODULE_ALIAS_BLOCKDEV_MAJOR(SCSI_DISK15_MAJOR);
94 MODULE_ALIAS_SCSI_DEVICE(TYPE_DISK);
95 MODULE_ALIAS_SCSI_DEVICE(TYPE_MOD);
96 MODULE_ALIAS_SCSI_DEVICE(TYPE_RBC);
97 MODULE_ALIAS_SCSI_DEVICE(TYPE_ZBC);
99 #if !defined(CONFIG_DEBUG_BLOCK_EXT_DEVT)
105 static void sd_config_discard(struct scsi_disk *, unsigned int);
106 static void sd_config_write_same(struct scsi_disk *);
107 static int sd_revalidate_disk(struct gendisk *);
108 static void sd_unlock_native_capacity(struct gendisk *disk);
109 static int sd_probe(struct device *);
110 static int sd_remove(struct device *);
111 static void sd_shutdown(struct device *);
112 static int sd_suspend_system(struct device *);
113 static int sd_suspend_runtime(struct device *);
114 static int sd_resume(struct device *);
115 static void sd_rescan(struct device *);
116 static int sd_init_command(struct scsi_cmnd *SCpnt);
117 static void sd_uninit_command(struct scsi_cmnd *SCpnt);
118 static int sd_done(struct scsi_cmnd *);
119 static void sd_eh_reset(struct scsi_cmnd *);
120 static int sd_eh_action(struct scsi_cmnd *, int);
121 static void sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer);
122 static void scsi_disk_release(struct device *cdev);
123 static void sd_print_sense_hdr(struct scsi_disk *, struct scsi_sense_hdr *);
124 static void sd_print_result(const struct scsi_disk *, const char *, int);
126 static DEFINE_SPINLOCK(sd_index_lock);
127 static DEFINE_IDA(sd_index_ida);
129 /* This semaphore is used to mediate the 0->1 reference get in the
130 * face of object destruction (i.e. we can't allow a get on an
131 * object after last put) */
132 static DEFINE_MUTEX(sd_ref_mutex);
134 static struct kmem_cache *sd_cdb_cache;
135 static mempool_t *sd_cdb_pool;
137 static const char *sd_cache_types[] = {
138 "write through", "none", "write back",
139 "write back, no read (daft)"
142 static void sd_set_flush_flag(struct scsi_disk *sdkp)
144 bool wc = false, fua = false;
152 blk_queue_write_cache(sdkp->disk->queue, wc, fua);
156 cache_type_store(struct device *dev, struct device_attribute *attr,
157 const char *buf, size_t count)
159 int ct, rcd, wce, sp;
160 struct scsi_disk *sdkp = to_scsi_disk(dev);
161 struct scsi_device *sdp = sdkp->device;
164 struct scsi_mode_data data;
165 struct scsi_sense_hdr sshdr;
166 static const char temp[] = "temporary ";
169 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
170 /* no cache control on RBC devices; theoretically they
171 * can do it, but there's probably so many exceptions
172 * it's not worth the risk */
175 if (strncmp(buf, temp, sizeof(temp) - 1) == 0) {
176 buf += sizeof(temp) - 1;
177 sdkp->cache_override = 1;
179 sdkp->cache_override = 0;
182 ct = sysfs_match_string(sd_cache_types, buf);
186 rcd = ct & 0x01 ? 1 : 0;
187 wce = (ct & 0x02) && !sdkp->write_prot ? 1 : 0;
189 if (sdkp->cache_override) {
192 sd_set_flush_flag(sdkp);
196 if (scsi_mode_sense(sdp, 0x08, 8, buffer, sizeof(buffer), SD_TIMEOUT,
197 SD_MAX_RETRIES, &data, NULL))
199 len = min_t(size_t, sizeof(buffer), data.length - data.header_length -
200 data.block_descriptor_length);
201 buffer_data = buffer + data.header_length +
202 data.block_descriptor_length;
203 buffer_data[2] &= ~0x05;
204 buffer_data[2] |= wce << 2 | rcd;
205 sp = buffer_data[0] & 0x80 ? 1 : 0;
206 buffer_data[0] &= ~0x80;
208 if (scsi_mode_select(sdp, 1, sp, 8, buffer_data, len, SD_TIMEOUT,
209 SD_MAX_RETRIES, &data, &sshdr)) {
210 if (scsi_sense_valid(&sshdr))
211 sd_print_sense_hdr(sdkp, &sshdr);
214 revalidate_disk(sdkp->disk);
219 manage_start_stop_show(struct device *dev, struct device_attribute *attr,
222 struct scsi_disk *sdkp = to_scsi_disk(dev);
223 struct scsi_device *sdp = sdkp->device;
225 return sprintf(buf, "%u\n", sdp->manage_start_stop);
229 manage_start_stop_store(struct device *dev, struct device_attribute *attr,
230 const char *buf, size_t count)
232 struct scsi_disk *sdkp = to_scsi_disk(dev);
233 struct scsi_device *sdp = sdkp->device;
235 if (!capable(CAP_SYS_ADMIN))
238 sdp->manage_start_stop = simple_strtoul(buf, NULL, 10);
242 static DEVICE_ATTR_RW(manage_start_stop);
245 allow_restart_show(struct device *dev, struct device_attribute *attr, char *buf)
247 struct scsi_disk *sdkp = to_scsi_disk(dev);
249 return sprintf(buf, "%u\n", sdkp->device->allow_restart);
253 allow_restart_store(struct device *dev, struct device_attribute *attr,
254 const char *buf, size_t count)
256 struct scsi_disk *sdkp = to_scsi_disk(dev);
257 struct scsi_device *sdp = sdkp->device;
259 if (!capable(CAP_SYS_ADMIN))
262 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
265 sdp->allow_restart = simple_strtoul(buf, NULL, 10);
269 static DEVICE_ATTR_RW(allow_restart);
272 cache_type_show(struct device *dev, struct device_attribute *attr, char *buf)
274 struct scsi_disk *sdkp = to_scsi_disk(dev);
275 int ct = sdkp->RCD + 2*sdkp->WCE;
277 return sprintf(buf, "%s\n", sd_cache_types[ct]);
279 static DEVICE_ATTR_RW(cache_type);
282 FUA_show(struct device *dev, struct device_attribute *attr, char *buf)
284 struct scsi_disk *sdkp = to_scsi_disk(dev);
286 return sprintf(buf, "%u\n", sdkp->DPOFUA);
288 static DEVICE_ATTR_RO(FUA);
291 protection_type_show(struct device *dev, struct device_attribute *attr,
294 struct scsi_disk *sdkp = to_scsi_disk(dev);
296 return sprintf(buf, "%u\n", sdkp->protection_type);
300 protection_type_store(struct device *dev, struct device_attribute *attr,
301 const char *buf, size_t count)
303 struct scsi_disk *sdkp = to_scsi_disk(dev);
307 if (!capable(CAP_SYS_ADMIN))
310 err = kstrtouint(buf, 10, &val);
315 if (val <= T10_PI_TYPE3_PROTECTION)
316 sdkp->protection_type = val;
320 static DEVICE_ATTR_RW(protection_type);
323 protection_mode_show(struct device *dev, struct device_attribute *attr,
326 struct scsi_disk *sdkp = to_scsi_disk(dev);
327 struct scsi_device *sdp = sdkp->device;
328 unsigned int dif, dix;
330 dif = scsi_host_dif_capable(sdp->host, sdkp->protection_type);
331 dix = scsi_host_dix_capable(sdp->host, sdkp->protection_type);
333 if (!dix && scsi_host_dix_capable(sdp->host, T10_PI_TYPE0_PROTECTION)) {
339 return sprintf(buf, "none\n");
341 return sprintf(buf, "%s%u\n", dix ? "dix" : "dif", dif);
343 static DEVICE_ATTR_RO(protection_mode);
346 app_tag_own_show(struct device *dev, struct device_attribute *attr, char *buf)
348 struct scsi_disk *sdkp = to_scsi_disk(dev);
350 return sprintf(buf, "%u\n", sdkp->ATO);
352 static DEVICE_ATTR_RO(app_tag_own);
355 thin_provisioning_show(struct device *dev, struct device_attribute *attr,
358 struct scsi_disk *sdkp = to_scsi_disk(dev);
360 return sprintf(buf, "%u\n", sdkp->lbpme);
362 static DEVICE_ATTR_RO(thin_provisioning);
364 /* sysfs_match_string() requires dense arrays */
365 static const char *lbp_mode[] = {
366 [SD_LBP_FULL] = "full",
367 [SD_LBP_UNMAP] = "unmap",
368 [SD_LBP_WS16] = "writesame_16",
369 [SD_LBP_WS10] = "writesame_10",
370 [SD_LBP_ZERO] = "writesame_zero",
371 [SD_LBP_DISABLE] = "disabled",
375 provisioning_mode_show(struct device *dev, struct device_attribute *attr,
378 struct scsi_disk *sdkp = to_scsi_disk(dev);
380 return sprintf(buf, "%s\n", lbp_mode[sdkp->provisioning_mode]);
384 provisioning_mode_store(struct device *dev, struct device_attribute *attr,
385 const char *buf, size_t count)
387 struct scsi_disk *sdkp = to_scsi_disk(dev);
388 struct scsi_device *sdp = sdkp->device;
391 if (!capable(CAP_SYS_ADMIN))
394 if (sd_is_zoned(sdkp)) {
395 sd_config_discard(sdkp, SD_LBP_DISABLE);
399 if (sdp->type != TYPE_DISK)
402 mode = sysfs_match_string(lbp_mode, buf);
406 sd_config_discard(sdkp, mode);
410 static DEVICE_ATTR_RW(provisioning_mode);
412 /* sysfs_match_string() requires dense arrays */
413 static const char *zeroing_mode[] = {
414 [SD_ZERO_WRITE] = "write",
415 [SD_ZERO_WS] = "writesame",
416 [SD_ZERO_WS16_UNMAP] = "writesame_16_unmap",
417 [SD_ZERO_WS10_UNMAP] = "writesame_10_unmap",
421 zeroing_mode_show(struct device *dev, struct device_attribute *attr,
424 struct scsi_disk *sdkp = to_scsi_disk(dev);
426 return sprintf(buf, "%s\n", zeroing_mode[sdkp->zeroing_mode]);
430 zeroing_mode_store(struct device *dev, struct device_attribute *attr,
431 const char *buf, size_t count)
433 struct scsi_disk *sdkp = to_scsi_disk(dev);
436 if (!capable(CAP_SYS_ADMIN))
439 mode = sysfs_match_string(zeroing_mode, buf);
443 sdkp->zeroing_mode = mode;
447 static DEVICE_ATTR_RW(zeroing_mode);
450 max_medium_access_timeouts_show(struct device *dev,
451 struct device_attribute *attr, char *buf)
453 struct scsi_disk *sdkp = to_scsi_disk(dev);
455 return sprintf(buf, "%u\n", sdkp->max_medium_access_timeouts);
459 max_medium_access_timeouts_store(struct device *dev,
460 struct device_attribute *attr, const char *buf,
463 struct scsi_disk *sdkp = to_scsi_disk(dev);
466 if (!capable(CAP_SYS_ADMIN))
469 err = kstrtouint(buf, 10, &sdkp->max_medium_access_timeouts);
471 return err ? err : count;
473 static DEVICE_ATTR_RW(max_medium_access_timeouts);
476 max_write_same_blocks_show(struct device *dev, struct device_attribute *attr,
479 struct scsi_disk *sdkp = to_scsi_disk(dev);
481 return sprintf(buf, "%u\n", sdkp->max_ws_blocks);
485 max_write_same_blocks_store(struct device *dev, struct device_attribute *attr,
486 const char *buf, size_t count)
488 struct scsi_disk *sdkp = to_scsi_disk(dev);
489 struct scsi_device *sdp = sdkp->device;
493 if (!capable(CAP_SYS_ADMIN))
496 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
499 err = kstrtoul(buf, 10, &max);
505 sdp->no_write_same = 1;
506 else if (max <= SD_MAX_WS16_BLOCKS) {
507 sdp->no_write_same = 0;
508 sdkp->max_ws_blocks = max;
511 sd_config_write_same(sdkp);
515 static DEVICE_ATTR_RW(max_write_same_blocks);
517 static struct attribute *sd_disk_attrs[] = {
518 &dev_attr_cache_type.attr,
520 &dev_attr_allow_restart.attr,
521 &dev_attr_manage_start_stop.attr,
522 &dev_attr_protection_type.attr,
523 &dev_attr_protection_mode.attr,
524 &dev_attr_app_tag_own.attr,
525 &dev_attr_thin_provisioning.attr,
526 &dev_attr_provisioning_mode.attr,
527 &dev_attr_zeroing_mode.attr,
528 &dev_attr_max_write_same_blocks.attr,
529 &dev_attr_max_medium_access_timeouts.attr,
532 ATTRIBUTE_GROUPS(sd_disk);
534 static struct class sd_disk_class = {
536 .owner = THIS_MODULE,
537 .dev_release = scsi_disk_release,
538 .dev_groups = sd_disk_groups,
541 static const struct dev_pm_ops sd_pm_ops = {
542 .suspend = sd_suspend_system,
544 .poweroff = sd_suspend_system,
545 .restore = sd_resume,
546 .runtime_suspend = sd_suspend_runtime,
547 .runtime_resume = sd_resume,
550 static struct scsi_driver sd_template = {
553 .owner = THIS_MODULE,
556 .shutdown = sd_shutdown,
560 .init_command = sd_init_command,
561 .uninit_command = sd_uninit_command,
563 .eh_action = sd_eh_action,
564 .eh_reset = sd_eh_reset,
568 * Dummy kobj_map->probe function.
569 * The default ->probe function will call modprobe, which is
570 * pointless as this module is already loaded.
572 static struct kobject *sd_default_probe(dev_t devt, int *partno, void *data)
578 * Device no to disk mapping:
580 * major disc2 disc p1
581 * |............|.............|....|....| <- dev_t
584 * Inside a major, we have 16k disks, however mapped non-
585 * contiguously. The first 16 disks are for major0, the next
586 * ones with major1, ... Disk 256 is for major0 again, disk 272
588 * As we stay compatible with our numbering scheme, we can reuse
589 * the well-know SCSI majors 8, 65--71, 136--143.
591 static int sd_major(int major_idx)
595 return SCSI_DISK0_MAJOR;
597 return SCSI_DISK1_MAJOR + major_idx - 1;
599 return SCSI_DISK8_MAJOR + major_idx - 8;
602 return 0; /* shut up gcc */
606 static struct scsi_disk *scsi_disk_get(struct gendisk *disk)
608 struct scsi_disk *sdkp = NULL;
610 mutex_lock(&sd_ref_mutex);
612 if (disk->private_data) {
613 sdkp = scsi_disk(disk);
614 if (scsi_device_get(sdkp->device) == 0)
615 get_device(&sdkp->dev);
619 mutex_unlock(&sd_ref_mutex);
623 static void scsi_disk_put(struct scsi_disk *sdkp)
625 struct scsi_device *sdev = sdkp->device;
627 mutex_lock(&sd_ref_mutex);
628 put_device(&sdkp->dev);
629 scsi_device_put(sdev);
630 mutex_unlock(&sd_ref_mutex);
633 #ifdef CONFIG_BLK_SED_OPAL
634 static int sd_sec_submit(void *data, u16 spsp, u8 secp, void *buffer,
635 size_t len, bool send)
637 struct scsi_device *sdev = data;
641 cdb[0] = send ? SECURITY_PROTOCOL_OUT : SECURITY_PROTOCOL_IN;
643 put_unaligned_be16(spsp, &cdb[2]);
644 put_unaligned_be32(len, &cdb[6]);
646 ret = scsi_execute_req(sdev, cdb,
647 send ? DMA_TO_DEVICE : DMA_FROM_DEVICE,
648 buffer, len, NULL, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
649 return ret <= 0 ? ret : -EIO;
651 #endif /* CONFIG_BLK_SED_OPAL */
653 static unsigned char sd_setup_protect_cmnd(struct scsi_cmnd *scmd,
654 unsigned int dix, unsigned int dif)
656 struct bio *bio = scmd->request->bio;
657 unsigned int prot_op = sd_prot_op(rq_data_dir(scmd->request), dix, dif);
658 unsigned int protect = 0;
660 if (dix) { /* DIX Type 0, 1, 2, 3 */
661 if (bio_integrity_flagged(bio, BIP_IP_CHECKSUM))
662 scmd->prot_flags |= SCSI_PROT_IP_CHECKSUM;
664 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
665 scmd->prot_flags |= SCSI_PROT_GUARD_CHECK;
668 if (dif != T10_PI_TYPE3_PROTECTION) { /* DIX/DIF Type 0, 1, 2 */
669 scmd->prot_flags |= SCSI_PROT_REF_INCREMENT;
671 if (bio_integrity_flagged(bio, BIP_CTRL_NOCHECK) == false)
672 scmd->prot_flags |= SCSI_PROT_REF_CHECK;
675 if (dif) { /* DIX/DIF Type 1, 2, 3 */
676 scmd->prot_flags |= SCSI_PROT_TRANSFER_PI;
678 if (bio_integrity_flagged(bio, BIP_DISK_NOCHECK))
679 protect = 3 << 5; /* Disable target PI checking */
681 protect = 1 << 5; /* Enable target PI checking */
684 scsi_set_prot_op(scmd, prot_op);
685 scsi_set_prot_type(scmd, dif);
686 scmd->prot_flags &= sd_prot_flag_mask(prot_op);
691 static void sd_config_discard(struct scsi_disk *sdkp, unsigned int mode)
693 struct request_queue *q = sdkp->disk->queue;
694 unsigned int logical_block_size = sdkp->device->sector_size;
695 unsigned int max_blocks = 0;
697 q->limits.discard_alignment =
698 sdkp->unmap_alignment * logical_block_size;
699 q->limits.discard_granularity =
700 max(sdkp->physical_block_size,
701 sdkp->unmap_granularity * logical_block_size);
702 sdkp->provisioning_mode = mode;
708 blk_queue_max_discard_sectors(q, 0);
709 queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q);
713 max_blocks = min_not_zero(sdkp->max_unmap_blocks,
714 (u32)SD_MAX_WS16_BLOCKS);
718 max_blocks = min_not_zero(sdkp->max_ws_blocks,
719 (u32)SD_MAX_WS16_BLOCKS);
723 max_blocks = min_not_zero(sdkp->max_ws_blocks,
724 (u32)SD_MAX_WS10_BLOCKS);
728 max_blocks = min_not_zero(sdkp->max_ws_blocks,
729 (u32)SD_MAX_WS10_BLOCKS);
733 blk_queue_max_discard_sectors(q, max_blocks * (logical_block_size >> 9));
734 queue_flag_set_unlocked(QUEUE_FLAG_DISCARD, q);
737 static int sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
739 struct scsi_device *sdp = cmd->device;
740 struct request *rq = cmd->request;
741 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
742 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
743 unsigned int data_len = 24;
746 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
747 if (!rq->special_vec.bv_page)
748 return BLKPREP_DEFER;
749 rq->special_vec.bv_offset = 0;
750 rq->special_vec.bv_len = data_len;
751 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
754 cmd->cmnd[0] = UNMAP;
757 buf = page_address(rq->special_vec.bv_page);
758 put_unaligned_be16(6 + 16, &buf[0]);
759 put_unaligned_be16(16, &buf[2]);
760 put_unaligned_be64(sector, &buf[8]);
761 put_unaligned_be32(nr_sectors, &buf[16]);
763 cmd->allowed = SD_MAX_RETRIES;
764 cmd->transfersize = data_len;
765 rq->timeout = SD_TIMEOUT;
766 scsi_req(rq)->resid_len = data_len;
768 return scsi_init_io(cmd);
771 static int sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd, bool unmap)
773 struct scsi_device *sdp = cmd->device;
774 struct request *rq = cmd->request;
775 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
776 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
777 u32 data_len = sdp->sector_size;
779 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
780 if (!rq->special_vec.bv_page)
781 return BLKPREP_DEFER;
782 rq->special_vec.bv_offset = 0;
783 rq->special_vec.bv_len = data_len;
784 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
787 cmd->cmnd[0] = WRITE_SAME_16;
789 cmd->cmnd[1] = 0x8; /* UNMAP */
790 put_unaligned_be64(sector, &cmd->cmnd[2]);
791 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
793 cmd->allowed = SD_MAX_RETRIES;
794 cmd->transfersize = data_len;
795 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
796 scsi_req(rq)->resid_len = data_len;
798 return scsi_init_io(cmd);
801 static int sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd, bool unmap)
803 struct scsi_device *sdp = cmd->device;
804 struct request *rq = cmd->request;
805 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
806 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
807 u32 data_len = sdp->sector_size;
809 rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
810 if (!rq->special_vec.bv_page)
811 return BLKPREP_DEFER;
812 rq->special_vec.bv_offset = 0;
813 rq->special_vec.bv_len = data_len;
814 rq->rq_flags |= RQF_SPECIAL_PAYLOAD;
817 cmd->cmnd[0] = WRITE_SAME;
819 cmd->cmnd[1] = 0x8; /* UNMAP */
820 put_unaligned_be32(sector, &cmd->cmnd[2]);
821 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
823 cmd->allowed = SD_MAX_RETRIES;
824 cmd->transfersize = data_len;
825 rq->timeout = unmap ? SD_TIMEOUT : SD_WRITE_SAME_TIMEOUT;
826 scsi_req(rq)->resid_len = data_len;
828 return scsi_init_io(cmd);
831 static int sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
833 struct request *rq = cmd->request;
834 struct scsi_device *sdp = cmd->device;
835 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
836 u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
837 u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
840 if (!(rq->cmd_flags & REQ_NOUNMAP)) {
841 switch (sdkp->zeroing_mode) {
842 case SD_ZERO_WS16_UNMAP:
843 ret = sd_setup_write_same16_cmnd(cmd, true);
845 case SD_ZERO_WS10_UNMAP:
846 ret = sd_setup_write_same10_cmnd(cmd, true);
851 if (sdp->no_write_same)
852 return BLKPREP_INVALID;
854 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff)
855 ret = sd_setup_write_same16_cmnd(cmd, false);
857 ret = sd_setup_write_same10_cmnd(cmd, false);
860 if (sd_is_zoned(sdkp) && ret == BLKPREP_OK)
861 return sd_zbc_write_lock_zone(cmd);
866 static void sd_config_write_same(struct scsi_disk *sdkp)
868 struct request_queue *q = sdkp->disk->queue;
869 unsigned int logical_block_size = sdkp->device->sector_size;
871 if (sdkp->device->no_write_same) {
872 sdkp->max_ws_blocks = 0;
876 /* Some devices can not handle block counts above 0xffff despite
877 * supporting WRITE SAME(16). Consequently we default to 64k
878 * blocks per I/O unless the device explicitly advertises a
881 if (sdkp->max_ws_blocks > SD_MAX_WS10_BLOCKS)
882 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
883 (u32)SD_MAX_WS16_BLOCKS);
884 else if (sdkp->ws16 || sdkp->ws10 || sdkp->device->no_report_opcodes)
885 sdkp->max_ws_blocks = min_not_zero(sdkp->max_ws_blocks,
886 (u32)SD_MAX_WS10_BLOCKS);
888 sdkp->device->no_write_same = 1;
889 sdkp->max_ws_blocks = 0;
892 if (sdkp->lbprz && sdkp->lbpws)
893 sdkp->zeroing_mode = SD_ZERO_WS16_UNMAP;
894 else if (sdkp->lbprz && sdkp->lbpws10)
895 sdkp->zeroing_mode = SD_ZERO_WS10_UNMAP;
896 else if (sdkp->max_ws_blocks)
897 sdkp->zeroing_mode = SD_ZERO_WS;
899 sdkp->zeroing_mode = SD_ZERO_WRITE;
902 blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
903 (logical_block_size >> 9));
904 blk_queue_max_write_zeroes_sectors(q, sdkp->max_ws_blocks *
905 (logical_block_size >> 9));
909 * sd_setup_write_same_cmnd - write the same data to multiple blocks
910 * @cmd: command to prepare
912 * Will set up either WRITE SAME(10) or WRITE SAME(16) depending on
913 * the preference indicated by the target device.
915 static int sd_setup_write_same_cmnd(struct scsi_cmnd *cmd)
917 struct request *rq = cmd->request;
918 struct scsi_device *sdp = cmd->device;
919 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
920 struct bio *bio = rq->bio;
921 sector_t sector = blk_rq_pos(rq);
922 unsigned int nr_sectors = blk_rq_sectors(rq);
923 unsigned int nr_bytes = blk_rq_bytes(rq);
926 if (sdkp->device->no_write_same)
927 return BLKPREP_INVALID;
929 BUG_ON(bio_offset(bio) || bio_iovec(bio).bv_len != sdp->sector_size);
931 if (sd_is_zoned(sdkp)) {
932 ret = sd_zbc_write_lock_zone(cmd);
933 if (ret != BLKPREP_OK)
937 sector >>= ilog2(sdp->sector_size) - 9;
938 nr_sectors >>= ilog2(sdp->sector_size) - 9;
940 rq->timeout = SD_WRITE_SAME_TIMEOUT;
942 if (sdkp->ws16 || sector > 0xffffffff || nr_sectors > 0xffff) {
944 cmd->cmnd[0] = WRITE_SAME_16;
945 put_unaligned_be64(sector, &cmd->cmnd[2]);
946 put_unaligned_be32(nr_sectors, &cmd->cmnd[10]);
949 cmd->cmnd[0] = WRITE_SAME;
950 put_unaligned_be32(sector, &cmd->cmnd[2]);
951 put_unaligned_be16(nr_sectors, &cmd->cmnd[7]);
954 cmd->transfersize = sdp->sector_size;
955 cmd->allowed = SD_MAX_RETRIES;
958 * For WRITE SAME the data transferred via the DATA OUT buffer is
959 * different from the amount of data actually written to the target.
961 * We set up __data_len to the amount of data transferred via the
962 * DATA OUT buffer so that blk_rq_map_sg sets up the proper S/G list
963 * to transfer a single sector of data first, but then reset it to
964 * the amount of data to be written right after so that the I/O path
965 * knows how much to actually write.
967 rq->__data_len = sdp->sector_size;
968 ret = scsi_init_io(cmd);
969 rq->__data_len = nr_bytes;
971 if (sd_is_zoned(sdkp) && ret != BLKPREP_OK)
972 sd_zbc_write_unlock_zone(cmd);
977 static int sd_setup_flush_cmnd(struct scsi_cmnd *cmd)
979 struct request *rq = cmd->request;
981 /* flush requests don't perform I/O, zero the S/G table */
982 memset(&cmd->sdb, 0, sizeof(cmd->sdb));
984 cmd->cmnd[0] = SYNCHRONIZE_CACHE;
986 cmd->transfersize = 0;
987 cmd->allowed = SD_MAX_RETRIES;
989 rq->timeout = rq->q->rq_timeout * SD_FLUSH_TIMEOUT_MULTIPLIER;
993 static int sd_setup_read_write_cmnd(struct scsi_cmnd *SCpnt)
995 struct request *rq = SCpnt->request;
996 struct scsi_device *sdp = SCpnt->device;
997 struct gendisk *disk = rq->rq_disk;
998 struct scsi_disk *sdkp = scsi_disk(disk);
999 sector_t block = blk_rq_pos(rq);
1001 unsigned int this_count = blk_rq_sectors(rq);
1002 unsigned int dif, dix;
1003 bool zoned_write = sd_is_zoned(sdkp) && rq_data_dir(rq) == WRITE;
1005 unsigned char protect;
1008 ret = sd_zbc_write_lock_zone(SCpnt);
1009 if (ret != BLKPREP_OK)
1013 ret = scsi_init_io(SCpnt);
1014 if (ret != BLKPREP_OK)
1016 WARN_ON_ONCE(SCpnt != rq->special);
1018 /* from here on until we're complete, any goto out
1019 * is used for a killable error condition */
1023 scmd_printk(KERN_INFO, SCpnt,
1024 "%s: block=%llu, count=%d\n",
1025 __func__, (unsigned long long)block, this_count));
1027 if (!sdp || !scsi_device_online(sdp) ||
1028 block + blk_rq_sectors(rq) > get_capacity(disk)) {
1029 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1030 "Finishing %u sectors\n",
1031 blk_rq_sectors(rq)));
1032 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1033 "Retry with 0x%p\n", SCpnt));
1039 * quietly refuse to do anything to a changed disc until
1040 * the changed bit has been reset
1042 /* printk("SCSI disk has been changed or is not present. Prohibiting further I/O.\n"); */
1047 * Some SD card readers can't handle multi-sector accesses which touch
1048 * the last one or two hardware sectors. Split accesses as needed.
1050 threshold = get_capacity(disk) - SD_LAST_BUGGY_SECTORS *
1051 (sdp->sector_size / 512);
1053 if (unlikely(sdp->last_sector_bug && block + this_count > threshold)) {
1054 if (block < threshold) {
1055 /* Access up to the threshold but not beyond */
1056 this_count = threshold - block;
1058 /* Access only a single hardware sector */
1059 this_count = sdp->sector_size / 512;
1063 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt, "block=%llu\n",
1064 (unsigned long long)block));
1067 * If we have a 1K hardware sectorsize, prevent access to single
1068 * 512 byte sectors. In theory we could handle this - in fact
1069 * the scsi cdrom driver must be able to handle this because
1070 * we typically use 1K blocksizes, and cdroms typically have
1071 * 2K hardware sectorsizes. Of course, things are simpler
1072 * with the cdrom, since it is read-only. For performance
1073 * reasons, the filesystems should be able to handle this
1074 * and not force the scsi disk driver to use bounce buffers
1077 if (sdp->sector_size == 1024) {
1078 if ((block & 1) || (blk_rq_sectors(rq) & 1)) {
1079 scmd_printk(KERN_ERR, SCpnt,
1080 "Bad block number requested\n");
1084 this_count = this_count >> 1;
1087 if (sdp->sector_size == 2048) {
1088 if ((block & 3) || (blk_rq_sectors(rq) & 3)) {
1089 scmd_printk(KERN_ERR, SCpnt,
1090 "Bad block number requested\n");
1094 this_count = this_count >> 2;
1097 if (sdp->sector_size == 4096) {
1098 if ((block & 7) || (blk_rq_sectors(rq) & 7)) {
1099 scmd_printk(KERN_ERR, SCpnt,
1100 "Bad block number requested\n");
1104 this_count = this_count >> 3;
1107 if (rq_data_dir(rq) == WRITE) {
1108 SCpnt->cmnd[0] = WRITE_6;
1110 if (blk_integrity_rq(rq))
1111 sd_dif_prepare(SCpnt);
1113 } else if (rq_data_dir(rq) == READ) {
1114 SCpnt->cmnd[0] = READ_6;
1116 scmd_printk(KERN_ERR, SCpnt, "Unknown command %d\n", req_op(rq));
1120 SCSI_LOG_HLQUEUE(2, scmd_printk(KERN_INFO, SCpnt,
1121 "%s %d/%u 512 byte blocks.\n",
1122 (rq_data_dir(rq) == WRITE) ?
1123 "writing" : "reading", this_count,
1124 blk_rq_sectors(rq)));
1126 dix = scsi_prot_sg_count(SCpnt);
1127 dif = scsi_host_dif_capable(SCpnt->device->host, sdkp->protection_type);
1130 protect = sd_setup_protect_cmnd(SCpnt, dix, dif);
1134 if (protect && sdkp->protection_type == T10_PI_TYPE2_PROTECTION) {
1135 SCpnt->cmnd = mempool_alloc(sd_cdb_pool, GFP_ATOMIC);
1137 if (unlikely(SCpnt->cmnd == NULL)) {
1138 ret = BLKPREP_DEFER;
1142 SCpnt->cmd_len = SD_EXT_CDB_SIZE;
1143 memset(SCpnt->cmnd, 0, SCpnt->cmd_len);
1144 SCpnt->cmnd[0] = VARIABLE_LENGTH_CMD;
1145 SCpnt->cmnd[7] = 0x18;
1146 SCpnt->cmnd[9] = (rq_data_dir(rq) == READ) ? READ_32 : WRITE_32;
1147 SCpnt->cmnd[10] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1150 SCpnt->cmnd[12] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1151 SCpnt->cmnd[13] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1152 SCpnt->cmnd[14] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1153 SCpnt->cmnd[15] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1154 SCpnt->cmnd[16] = (unsigned char) (block >> 24) & 0xff;
1155 SCpnt->cmnd[17] = (unsigned char) (block >> 16) & 0xff;
1156 SCpnt->cmnd[18] = (unsigned char) (block >> 8) & 0xff;
1157 SCpnt->cmnd[19] = (unsigned char) block & 0xff;
1159 /* Expected Indirect LBA */
1160 SCpnt->cmnd[20] = (unsigned char) (block >> 24) & 0xff;
1161 SCpnt->cmnd[21] = (unsigned char) (block >> 16) & 0xff;
1162 SCpnt->cmnd[22] = (unsigned char) (block >> 8) & 0xff;
1163 SCpnt->cmnd[23] = (unsigned char) block & 0xff;
1165 /* Transfer length */
1166 SCpnt->cmnd[28] = (unsigned char) (this_count >> 24) & 0xff;
1167 SCpnt->cmnd[29] = (unsigned char) (this_count >> 16) & 0xff;
1168 SCpnt->cmnd[30] = (unsigned char) (this_count >> 8) & 0xff;
1169 SCpnt->cmnd[31] = (unsigned char) this_count & 0xff;
1170 } else if (sdp->use_16_for_rw || (this_count > 0xffff)) {
1171 SCpnt->cmnd[0] += READ_16 - READ_6;
1172 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1173 SCpnt->cmnd[2] = sizeof(block) > 4 ? (unsigned char) (block >> 56) & 0xff : 0;
1174 SCpnt->cmnd[3] = sizeof(block) > 4 ? (unsigned char) (block >> 48) & 0xff : 0;
1175 SCpnt->cmnd[4] = sizeof(block) > 4 ? (unsigned char) (block >> 40) & 0xff : 0;
1176 SCpnt->cmnd[5] = sizeof(block) > 4 ? (unsigned char) (block >> 32) & 0xff : 0;
1177 SCpnt->cmnd[6] = (unsigned char) (block >> 24) & 0xff;
1178 SCpnt->cmnd[7] = (unsigned char) (block >> 16) & 0xff;
1179 SCpnt->cmnd[8] = (unsigned char) (block >> 8) & 0xff;
1180 SCpnt->cmnd[9] = (unsigned char) block & 0xff;
1181 SCpnt->cmnd[10] = (unsigned char) (this_count >> 24) & 0xff;
1182 SCpnt->cmnd[11] = (unsigned char) (this_count >> 16) & 0xff;
1183 SCpnt->cmnd[12] = (unsigned char) (this_count >> 8) & 0xff;
1184 SCpnt->cmnd[13] = (unsigned char) this_count & 0xff;
1185 SCpnt->cmnd[14] = SCpnt->cmnd[15] = 0;
1186 } else if ((this_count > 0xff) || (block > 0x1fffff) ||
1187 scsi_device_protection(SCpnt->device) ||
1188 SCpnt->device->use_10_for_rw) {
1189 SCpnt->cmnd[0] += READ_10 - READ_6;
1190 SCpnt->cmnd[1] = protect | ((rq->cmd_flags & REQ_FUA) ? 0x8 : 0);
1191 SCpnt->cmnd[2] = (unsigned char) (block >> 24) & 0xff;
1192 SCpnt->cmnd[3] = (unsigned char) (block >> 16) & 0xff;
1193 SCpnt->cmnd[4] = (unsigned char) (block >> 8) & 0xff;
1194 SCpnt->cmnd[5] = (unsigned char) block & 0xff;
1195 SCpnt->cmnd[6] = SCpnt->cmnd[9] = 0;
1196 SCpnt->cmnd[7] = (unsigned char) (this_count >> 8) & 0xff;
1197 SCpnt->cmnd[8] = (unsigned char) this_count & 0xff;
1199 if (unlikely(rq->cmd_flags & REQ_FUA)) {
1201 * This happens only if this drive failed
1202 * 10byte rw command with ILLEGAL_REQUEST
1203 * during operation and thus turned off
1206 scmd_printk(KERN_ERR, SCpnt,
1207 "FUA write on READ/WRITE(6) drive\n");
1211 SCpnt->cmnd[1] |= (unsigned char) ((block >> 16) & 0x1f);
1212 SCpnt->cmnd[2] = (unsigned char) ((block >> 8) & 0xff);
1213 SCpnt->cmnd[3] = (unsigned char) block & 0xff;
1214 SCpnt->cmnd[4] = (unsigned char) this_count;
1217 SCpnt->sdb.length = this_count * sdp->sector_size;
1220 * We shouldn't disconnect in the middle of a sector, so with a dumb
1221 * host adapter, it's safe to assume that we can at least transfer
1222 * this many bytes between each connect / disconnect.
1224 SCpnt->transfersize = sdp->sector_size;
1225 SCpnt->underflow = this_count << 9;
1226 SCpnt->allowed = SD_MAX_RETRIES;
1229 * This indicates that the command is ready from our end to be
1234 if (zoned_write && ret != BLKPREP_OK)
1235 sd_zbc_write_unlock_zone(SCpnt);
1240 static int sd_init_command(struct scsi_cmnd *cmd)
1242 struct request *rq = cmd->request;
1244 switch (req_op(rq)) {
1245 case REQ_OP_DISCARD:
1246 switch (scsi_disk(rq->rq_disk)->provisioning_mode) {
1248 return sd_setup_unmap_cmnd(cmd);
1250 return sd_setup_write_same16_cmnd(cmd, true);
1252 return sd_setup_write_same10_cmnd(cmd, true);
1254 return sd_setup_write_same10_cmnd(cmd, false);
1256 return BLKPREP_INVALID;
1258 case REQ_OP_WRITE_ZEROES:
1259 return sd_setup_write_zeroes_cmnd(cmd);
1260 case REQ_OP_WRITE_SAME:
1261 return sd_setup_write_same_cmnd(cmd);
1263 return sd_setup_flush_cmnd(cmd);
1266 return sd_setup_read_write_cmnd(cmd);
1267 case REQ_OP_ZONE_REPORT:
1268 return sd_zbc_setup_report_cmnd(cmd);
1269 case REQ_OP_ZONE_RESET:
1270 return sd_zbc_setup_reset_cmnd(cmd);
1276 static void sd_uninit_command(struct scsi_cmnd *SCpnt)
1278 struct request *rq = SCpnt->request;
1280 if (SCpnt->flags & SCMD_ZONE_WRITE_LOCK)
1281 sd_zbc_write_unlock_zone(SCpnt);
1283 if (rq->rq_flags & RQF_SPECIAL_PAYLOAD)
1284 __free_page(rq->special_vec.bv_page);
1286 if (SCpnt->cmnd != scsi_req(rq)->cmd) {
1287 mempool_free(SCpnt->cmnd, sd_cdb_pool);
1294 * sd_open - open a scsi disk device
1295 * @bdev: Block device of the scsi disk to open
1296 * @mode: FMODE_* mask
1298 * Returns 0 if successful. Returns a negated errno value in case
1301 * Note: This can be called from a user context (e.g. fsck(1) )
1302 * or from within the kernel (e.g. as a result of a mount(1) ).
1303 * In the latter case @inode and @filp carry an abridged amount
1304 * of information as noted above.
1306 * Locking: called with bdev->bd_mutex held.
1308 static int sd_open(struct block_device *bdev, fmode_t mode)
1310 struct scsi_disk *sdkp = scsi_disk_get(bdev->bd_disk);
1311 struct scsi_device *sdev;
1317 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_open\n"));
1319 sdev = sdkp->device;
1322 * If the device is in error recovery, wait until it is done.
1323 * If the device is offline, then disallow any access to it.
1326 if (!scsi_block_when_processing_errors(sdev))
1329 if (sdev->removable || sdkp->write_prot)
1330 check_disk_change(bdev);
1333 * If the drive is empty, just let the open fail.
1335 retval = -ENOMEDIUM;
1336 if (sdev->removable && !sdkp->media_present && !(mode & FMODE_NDELAY))
1340 * If the device has the write protect tab set, have the open fail
1341 * if the user expects to be able to write to the thing.
1344 if (sdkp->write_prot && (mode & FMODE_WRITE))
1348 * It is possible that the disk changing stuff resulted in
1349 * the device being taken offline. If this is the case,
1350 * report this to the user, and don't pretend that the
1351 * open actually succeeded.
1354 if (!scsi_device_online(sdev))
1357 if ((atomic_inc_return(&sdkp->openers) == 1) && sdev->removable) {
1358 if (scsi_block_when_processing_errors(sdev))
1359 scsi_set_medium_removal(sdev, SCSI_REMOVAL_PREVENT);
1365 scsi_disk_put(sdkp);
1370 * sd_release - invoked when the (last) close(2) is called on this
1372 * @disk: disk to release
1373 * @mode: FMODE_* mask
1377 * Note: may block (uninterruptible) if error recovery is underway
1380 * Locking: called with bdev->bd_mutex held.
1382 static void sd_release(struct gendisk *disk, fmode_t mode)
1384 struct scsi_disk *sdkp = scsi_disk(disk);
1385 struct scsi_device *sdev = sdkp->device;
1387 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_release\n"));
1389 if (atomic_dec_return(&sdkp->openers) == 0 && sdev->removable) {
1390 if (scsi_block_when_processing_errors(sdev))
1391 scsi_set_medium_removal(sdev, SCSI_REMOVAL_ALLOW);
1395 * XXX and what if there are packets in flight and this close()
1396 * XXX is followed by a "rmmod sd_mod"?
1399 scsi_disk_put(sdkp);
1402 static int sd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
1404 struct scsi_disk *sdkp = scsi_disk(bdev->bd_disk);
1405 struct scsi_device *sdp = sdkp->device;
1406 struct Scsi_Host *host = sdp->host;
1407 sector_t capacity = logical_to_sectors(sdp, sdkp->capacity);
1410 /* default to most commonly used values */
1411 diskinfo[0] = 0x40; /* 1 << 6 */
1412 diskinfo[1] = 0x20; /* 1 << 5 */
1413 diskinfo[2] = capacity >> 11;
1415 /* override with calculated, extended default, or driver values */
1416 if (host->hostt->bios_param)
1417 host->hostt->bios_param(sdp, bdev, capacity, diskinfo);
1419 scsicam_bios_param(bdev, capacity, diskinfo);
1421 geo->heads = diskinfo[0];
1422 geo->sectors = diskinfo[1];
1423 geo->cylinders = diskinfo[2];
1428 * sd_ioctl - process an ioctl
1429 * @bdev: target block device
1430 * @mode: FMODE_* mask
1431 * @cmd: ioctl command number
1432 * @arg: this is third argument given to ioctl(2) system call.
1433 * Often contains a pointer.
1435 * Returns 0 if successful (some ioctls return positive numbers on
1436 * success as well). Returns a negated errno value in case of error.
1438 * Note: most ioctls are forward onto the block subsystem or further
1439 * down in the scsi subsystem.
1441 static int sd_ioctl(struct block_device *bdev, fmode_t mode,
1442 unsigned int cmd, unsigned long arg)
1444 struct gendisk *disk = bdev->bd_disk;
1445 struct scsi_disk *sdkp = scsi_disk(disk);
1446 struct scsi_device *sdp = sdkp->device;
1447 void __user *p = (void __user *)arg;
1450 SCSI_LOG_IOCTL(1, sd_printk(KERN_INFO, sdkp, "sd_ioctl: disk=%s, "
1451 "cmd=0x%x\n", disk->disk_name, cmd));
1453 error = scsi_verify_blk_ioctl(bdev, cmd);
1458 * If we are in the middle of error recovery, don't let anyone
1459 * else try and use this device. Also, if error recovery fails, it
1460 * may try and take the device offline, in which case all further
1461 * access to the device is prohibited.
1463 error = scsi_ioctl_block_when_processing_errors(sdp, cmd,
1464 (mode & FMODE_NDELAY) != 0);
1468 if (is_sed_ioctl(cmd))
1469 return sed_ioctl(sdkp->opal_dev, cmd, p);
1472 * Send SCSI addressing ioctls directly to mid level, send other
1473 * ioctls to block level and then onto mid level if they can't be
1477 case SCSI_IOCTL_GET_IDLUN:
1478 case SCSI_IOCTL_GET_BUS_NUMBER:
1479 error = scsi_ioctl(sdp, cmd, p);
1482 error = scsi_cmd_blk_ioctl(bdev, mode, cmd, p);
1483 if (error != -ENOTTY)
1485 error = scsi_ioctl(sdp, cmd, p);
1492 static void set_media_not_present(struct scsi_disk *sdkp)
1494 if (sdkp->media_present)
1495 sdkp->device->changed = 1;
1497 if (sdkp->device->removable) {
1498 sdkp->media_present = 0;
1503 static int media_not_present(struct scsi_disk *sdkp,
1504 struct scsi_sense_hdr *sshdr)
1506 if (!scsi_sense_valid(sshdr))
1509 /* not invoked for commands that could return deferred errors */
1510 switch (sshdr->sense_key) {
1511 case UNIT_ATTENTION:
1513 /* medium not present */
1514 if (sshdr->asc == 0x3A) {
1515 set_media_not_present(sdkp);
1523 * sd_check_events - check media events
1524 * @disk: kernel device descriptor
1525 * @clearing: disk events currently being cleared
1527 * Returns mask of DISK_EVENT_*.
1529 * Note: this function is invoked from the block subsystem.
1531 static unsigned int sd_check_events(struct gendisk *disk, unsigned int clearing)
1533 struct scsi_disk *sdkp = scsi_disk_get(disk);
1534 struct scsi_device *sdp;
1541 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp, "sd_check_events\n"));
1544 * If the device is offline, don't send any commands - just pretend as
1545 * if the command failed. If the device ever comes back online, we
1546 * can deal with it then. It is only because of unrecoverable errors
1547 * that we would ever take a device offline in the first place.
1549 if (!scsi_device_online(sdp)) {
1550 set_media_not_present(sdkp);
1555 * Using TEST_UNIT_READY enables differentiation between drive with
1556 * no cartridge loaded - NOT READY, drive with changed cartridge -
1557 * UNIT ATTENTION, or with same cartridge - GOOD STATUS.
1559 * Drives that auto spin down. eg iomega jaz 1G, will be started
1560 * by sd_spinup_disk() from sd_revalidate_disk(), which happens whenever
1561 * sd_revalidate() is called.
1563 if (scsi_block_when_processing_errors(sdp)) {
1564 struct scsi_sense_hdr sshdr = { 0, };
1566 retval = scsi_test_unit_ready(sdp, SD_TIMEOUT, SD_MAX_RETRIES,
1569 /* failed to execute TUR, assume media not present */
1570 if (host_byte(retval)) {
1571 set_media_not_present(sdkp);
1575 if (media_not_present(sdkp, &sshdr))
1580 * For removable scsi disk we have to recognise the presence
1581 * of a disk in the drive.
1583 if (!sdkp->media_present)
1585 sdkp->media_present = 1;
1588 * sdp->changed is set under the following conditions:
1590 * Medium present state has changed in either direction.
1591 * Device has indicated UNIT_ATTENTION.
1593 retval = sdp->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1595 scsi_disk_put(sdkp);
1599 static int sd_sync_cache(struct scsi_disk *sdkp, struct scsi_sense_hdr *sshdr)
1602 struct scsi_device *sdp = sdkp->device;
1603 const int timeout = sdp->request_queue->rq_timeout
1604 * SD_FLUSH_TIMEOUT_MULTIPLIER;
1605 struct scsi_sense_hdr my_sshdr;
1607 if (!scsi_device_online(sdp))
1610 /* caller might not be interested in sense, but we need it */
1614 for (retries = 3; retries > 0; --retries) {
1615 unsigned char cmd[10] = { 0 };
1617 cmd[0] = SYNCHRONIZE_CACHE;
1619 * Leave the rest of the command zero to indicate
1622 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, sshdr,
1623 timeout, SD_MAX_RETRIES, 0, RQF_PM, NULL);
1629 sd_print_result(sdkp, "Synchronize Cache(10) failed", res);
1631 if (driver_byte(res) & DRIVER_SENSE)
1632 sd_print_sense_hdr(sdkp, sshdr);
1634 /* we need to evaluate the error return */
1635 if (scsi_sense_valid(sshdr) &&
1636 (sshdr->asc == 0x3a || /* medium not present */
1637 sshdr->asc == 0x20)) /* invalid command */
1638 /* this is no error here */
1641 switch (host_byte(res)) {
1642 /* ignore errors due to racing a disconnection */
1643 case DID_BAD_TARGET:
1644 case DID_NO_CONNECT:
1646 /* signal the upper layer it might try again */
1650 case DID_SOFT_ERROR:
1659 static void sd_rescan(struct device *dev)
1661 struct scsi_disk *sdkp = dev_get_drvdata(dev);
1663 revalidate_disk(sdkp->disk);
1667 #ifdef CONFIG_COMPAT
1669 * This gets directly called from VFS. When the ioctl
1670 * is not recognized we go back to the other translation paths.
1672 static int sd_compat_ioctl(struct block_device *bdev, fmode_t mode,
1673 unsigned int cmd, unsigned long arg)
1675 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1678 error = scsi_ioctl_block_when_processing_errors(sdev, cmd,
1679 (mode & FMODE_NDELAY) != 0);
1684 * Let the static ioctl translation table take care of it.
1686 if (!sdev->host->hostt->compat_ioctl)
1687 return -ENOIOCTLCMD;
1688 return sdev->host->hostt->compat_ioctl(sdev, cmd, (void __user *)arg);
1692 static char sd_pr_type(enum pr_type type)
1695 case PR_WRITE_EXCLUSIVE:
1697 case PR_EXCLUSIVE_ACCESS:
1699 case PR_WRITE_EXCLUSIVE_REG_ONLY:
1701 case PR_EXCLUSIVE_ACCESS_REG_ONLY:
1703 case PR_WRITE_EXCLUSIVE_ALL_REGS:
1705 case PR_EXCLUSIVE_ACCESS_ALL_REGS:
1712 static int sd_pr_command(struct block_device *bdev, u8 sa,
1713 u64 key, u64 sa_key, u8 type, u8 flags)
1715 struct scsi_device *sdev = scsi_disk(bdev->bd_disk)->device;
1716 struct scsi_sense_hdr sshdr;
1718 u8 cmd[16] = { 0, };
1719 u8 data[24] = { 0, };
1721 cmd[0] = PERSISTENT_RESERVE_OUT;
1724 put_unaligned_be32(sizeof(data), &cmd[5]);
1726 put_unaligned_be64(key, &data[0]);
1727 put_unaligned_be64(sa_key, &data[8]);
1730 result = scsi_execute_req(sdev, cmd, DMA_TO_DEVICE, &data, sizeof(data),
1731 &sshdr, SD_TIMEOUT, SD_MAX_RETRIES, NULL);
1733 if ((driver_byte(result) & DRIVER_SENSE) &&
1734 (scsi_sense_valid(&sshdr))) {
1735 sdev_printk(KERN_INFO, sdev, "PR command failed: %d\n", result);
1736 scsi_print_sense_hdr(sdev, NULL, &sshdr);
1742 static int sd_pr_register(struct block_device *bdev, u64 old_key, u64 new_key,
1745 if (flags & ~PR_FL_IGNORE_KEY)
1747 return sd_pr_command(bdev, (flags & PR_FL_IGNORE_KEY) ? 0x06 : 0x00,
1748 old_key, new_key, 0,
1749 (1 << 0) /* APTPL */);
1752 static int sd_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type,
1757 return sd_pr_command(bdev, 0x01, key, 0, sd_pr_type(type), 0);
1760 static int sd_pr_release(struct block_device *bdev, u64 key, enum pr_type type)
1762 return sd_pr_command(bdev, 0x02, key, 0, sd_pr_type(type), 0);
1765 static int sd_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key,
1766 enum pr_type type, bool abort)
1768 return sd_pr_command(bdev, abort ? 0x05 : 0x04, old_key, new_key,
1769 sd_pr_type(type), 0);
1772 static int sd_pr_clear(struct block_device *bdev, u64 key)
1774 return sd_pr_command(bdev, 0x03, key, 0, 0, 0);
1777 static const struct pr_ops sd_pr_ops = {
1778 .pr_register = sd_pr_register,
1779 .pr_reserve = sd_pr_reserve,
1780 .pr_release = sd_pr_release,
1781 .pr_preempt = sd_pr_preempt,
1782 .pr_clear = sd_pr_clear,
1785 static const struct block_device_operations sd_fops = {
1786 .owner = THIS_MODULE,
1788 .release = sd_release,
1790 .getgeo = sd_getgeo,
1791 #ifdef CONFIG_COMPAT
1792 .compat_ioctl = sd_compat_ioctl,
1794 .check_events = sd_check_events,
1795 .revalidate_disk = sd_revalidate_disk,
1796 .unlock_native_capacity = sd_unlock_native_capacity,
1797 .pr_ops = &sd_pr_ops,
1801 * sd_eh_reset - reset error handling callback
1802 * @scmd: sd-issued command that has failed
1804 * This function is called by the SCSI midlayer before starting
1805 * SCSI EH. When counting medium access failures we have to be
1806 * careful to register it only only once per device and SCSI EH run;
1807 * there might be several timed out commands which will cause the
1808 * 'max_medium_access_timeouts' counter to trigger after the first
1809 * SCSI EH run already and set the device to offline.
1810 * So this function resets the internal counter before starting SCSI EH.
1812 static void sd_eh_reset(struct scsi_cmnd *scmd)
1814 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1816 /* New SCSI EH run, reset gate variable */
1817 sdkp->ignore_medium_access_errors = false;
1821 * sd_eh_action - error handling callback
1822 * @scmd: sd-issued command that has failed
1823 * @eh_disp: The recovery disposition suggested by the midlayer
1825 * This function is called by the SCSI midlayer upon completion of an
1826 * error test command (currently TEST UNIT READY). The result of sending
1827 * the eh command is passed in eh_disp. We're looking for devices that
1828 * fail medium access commands but are OK with non access commands like
1829 * test unit ready (so wrongly see the device as having a successful
1832 static int sd_eh_action(struct scsi_cmnd *scmd, int eh_disp)
1834 struct scsi_disk *sdkp = scsi_disk(scmd->request->rq_disk);
1835 struct scsi_device *sdev = scmd->device;
1837 if (!scsi_device_online(sdev) ||
1838 !scsi_medium_access_command(scmd) ||
1839 host_byte(scmd->result) != DID_TIME_OUT ||
1844 * The device has timed out executing a medium access command.
1845 * However, the TEST UNIT READY command sent during error
1846 * handling completed successfully. Either the device is in the
1847 * process of recovering or has it suffered an internal failure
1848 * that prevents access to the storage medium.
1850 if (!sdkp->ignore_medium_access_errors) {
1851 sdkp->medium_access_timed_out++;
1852 sdkp->ignore_medium_access_errors = true;
1856 * If the device keeps failing read/write commands but TEST UNIT
1857 * READY always completes successfully we assume that medium
1858 * access is no longer possible and take the device offline.
1860 if (sdkp->medium_access_timed_out >= sdkp->max_medium_access_timeouts) {
1861 scmd_printk(KERN_ERR, scmd,
1862 "Medium access timeout failure. Offlining disk!\n");
1863 mutex_lock(&sdev->state_mutex);
1864 scsi_device_set_state(sdev, SDEV_OFFLINE);
1865 mutex_unlock(&sdev->state_mutex);
1873 static unsigned int sd_completed_bytes(struct scsi_cmnd *scmd)
1875 struct request *req = scmd->request;
1876 struct scsi_device *sdev = scmd->device;
1877 unsigned int transferred, good_bytes;
1878 u64 start_lba, end_lba, bad_lba;
1881 * Some commands have a payload smaller than the device logical
1882 * block size (e.g. INQUIRY on a 4K disk).
1884 if (scsi_bufflen(scmd) <= sdev->sector_size)
1887 /* Check if we have a 'bad_lba' information */
1888 if (!scsi_get_sense_info_fld(scmd->sense_buffer,
1889 SCSI_SENSE_BUFFERSIZE,
1894 * If the bad lba was reported incorrectly, we have no idea where
1897 start_lba = sectors_to_logical(sdev, blk_rq_pos(req));
1898 end_lba = start_lba + bytes_to_logical(sdev, scsi_bufflen(scmd));
1899 if (bad_lba < start_lba || bad_lba >= end_lba)
1903 * resid is optional but mostly filled in. When it's unused,
1904 * its value is zero, so we assume the whole buffer transferred
1906 transferred = scsi_bufflen(scmd) - scsi_get_resid(scmd);
1908 /* This computation should always be done in terms of the
1909 * resolution of the device's medium.
1911 good_bytes = logical_to_bytes(sdev, bad_lba - start_lba);
1913 return min(good_bytes, transferred);
1917 * sd_done - bottom half handler: called when the lower level
1918 * driver has completed (successfully or otherwise) a scsi command.
1919 * @SCpnt: mid-level's per command structure.
1921 * Note: potentially run from within an ISR. Must not block.
1923 static int sd_done(struct scsi_cmnd *SCpnt)
1925 int result = SCpnt->result;
1926 unsigned int good_bytes = result ? 0 : scsi_bufflen(SCpnt);
1927 unsigned int sector_size = SCpnt->device->sector_size;
1929 struct scsi_sense_hdr sshdr;
1930 struct scsi_disk *sdkp = scsi_disk(SCpnt->request->rq_disk);
1931 struct request *req = SCpnt->request;
1932 int sense_valid = 0;
1933 int sense_deferred = 0;
1935 switch (req_op(req)) {
1936 case REQ_OP_DISCARD:
1937 case REQ_OP_WRITE_ZEROES:
1938 case REQ_OP_WRITE_SAME:
1939 case REQ_OP_ZONE_RESET:
1941 good_bytes = blk_rq_bytes(req);
1942 scsi_set_resid(SCpnt, 0);
1945 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1948 case REQ_OP_ZONE_REPORT:
1950 good_bytes = scsi_bufflen(SCpnt)
1951 - scsi_get_resid(SCpnt);
1952 scsi_set_resid(SCpnt, 0);
1955 scsi_set_resid(SCpnt, blk_rq_bytes(req));
1960 * In case of bogus fw or device, we could end up having
1961 * an unaligned partial completion. Check this here and force
1964 resid = scsi_get_resid(SCpnt);
1965 if (resid & (sector_size - 1)) {
1966 sd_printk(KERN_INFO, sdkp,
1967 "Unaligned partial completion (resid=%u, sector_sz=%u)\n",
1968 resid, sector_size);
1969 resid = min(scsi_bufflen(SCpnt),
1970 round_up(resid, sector_size));
1971 scsi_set_resid(SCpnt, resid);
1976 sense_valid = scsi_command_normalize_sense(SCpnt, &sshdr);
1978 sense_deferred = scsi_sense_is_deferred(&sshdr);
1980 sdkp->medium_access_timed_out = 0;
1982 if (driver_byte(result) != DRIVER_SENSE &&
1983 (!sense_valid || sense_deferred))
1986 switch (sshdr.sense_key) {
1987 case HARDWARE_ERROR:
1989 good_bytes = sd_completed_bytes(SCpnt);
1991 case RECOVERED_ERROR:
1992 good_bytes = scsi_bufflen(SCpnt);
1995 /* This indicates a false check condition, so ignore it. An
1996 * unknown amount of data was transferred so treat it as an
2000 memset(SCpnt->sense_buffer, 0, SCSI_SENSE_BUFFERSIZE);
2002 case ABORTED_COMMAND:
2003 if (sshdr.asc == 0x10) /* DIF: Target detected corruption */
2004 good_bytes = sd_completed_bytes(SCpnt);
2006 case ILLEGAL_REQUEST:
2007 switch (sshdr.asc) {
2008 case 0x10: /* DIX: Host detected corruption */
2009 good_bytes = sd_completed_bytes(SCpnt);
2011 case 0x20: /* INVALID COMMAND OPCODE */
2012 case 0x24: /* INVALID FIELD IN CDB */
2013 switch (SCpnt->cmnd[0]) {
2015 sd_config_discard(sdkp, SD_LBP_DISABLE);
2019 if (SCpnt->cmnd[1] & 8) { /* UNMAP */
2020 sd_config_discard(sdkp, SD_LBP_DISABLE);
2022 sdkp->device->no_write_same = 1;
2023 sd_config_write_same(sdkp);
2024 req->__data_len = blk_rq_bytes(req);
2025 req->rq_flags |= RQF_QUIET;
2036 if (sd_is_zoned(sdkp))
2037 sd_zbc_complete(SCpnt, good_bytes, &sshdr);
2039 SCSI_LOG_HLCOMPLETE(1, scmd_printk(KERN_INFO, SCpnt,
2040 "sd_done: completed %d of %d bytes\n",
2041 good_bytes, scsi_bufflen(SCpnt)));
2043 if (rq_data_dir(SCpnt->request) == READ && scsi_prot_sg_count(SCpnt))
2044 sd_dif_complete(SCpnt, good_bytes);
2050 * spinup disk - called only in sd_revalidate_disk()
2053 sd_spinup_disk(struct scsi_disk *sdkp)
2055 unsigned char cmd[10];
2056 unsigned long spintime_expire = 0;
2057 int retries, spintime;
2058 unsigned int the_result;
2059 struct scsi_sense_hdr sshdr;
2060 int sense_valid = 0;
2064 /* Spin up drives, as required. Only do this at boot time */
2065 /* Spinup needs to be done for module loads too. */
2070 cmd[0] = TEST_UNIT_READY;
2071 memset((void *) &cmd[1], 0, 9);
2073 the_result = scsi_execute_req(sdkp->device, cmd,
2076 SD_MAX_RETRIES, NULL);
2079 * If the drive has indicated to us that it
2080 * doesn't have any media in it, don't bother
2081 * with any more polling.
2083 if (media_not_present(sdkp, &sshdr))
2087 sense_valid = scsi_sense_valid(&sshdr);
2089 } while (retries < 3 &&
2090 (!scsi_status_is_good(the_result) ||
2091 ((driver_byte(the_result) & DRIVER_SENSE) &&
2092 sense_valid && sshdr.sense_key == UNIT_ATTENTION)));
2094 if ((driver_byte(the_result) & DRIVER_SENSE) == 0) {
2095 /* no sense, TUR either succeeded or failed
2096 * with a status error */
2097 if(!spintime && !scsi_status_is_good(the_result)) {
2098 sd_print_result(sdkp, "Test Unit Ready failed",
2105 * The device does not want the automatic start to be issued.
2107 if (sdkp->device->no_start_on_add)
2110 if (sense_valid && sshdr.sense_key == NOT_READY) {
2111 if (sshdr.asc == 4 && sshdr.ascq == 3)
2112 break; /* manual intervention required */
2113 if (sshdr.asc == 4 && sshdr.ascq == 0xb)
2114 break; /* standby */
2115 if (sshdr.asc == 4 && sshdr.ascq == 0xc)
2116 break; /* unavailable */
2118 * Issue command to spin up drive when not ready
2121 sd_printk(KERN_NOTICE, sdkp, "Spinning up disk...");
2122 cmd[0] = START_STOP;
2123 cmd[1] = 1; /* Return immediately */
2124 memset((void *) &cmd[2], 0, 8);
2125 cmd[4] = 1; /* Start spin cycle */
2126 if (sdkp->device->start_stop_pwr_cond)
2128 scsi_execute_req(sdkp->device, cmd, DMA_NONE,
2130 SD_TIMEOUT, SD_MAX_RETRIES,
2132 spintime_expire = jiffies + 100 * HZ;
2135 /* Wait 1 second for next try */
2140 * Wait for USB flash devices with slow firmware.
2141 * Yes, this sense key/ASC combination shouldn't
2142 * occur here. It's characteristic of these devices.
2144 } else if (sense_valid &&
2145 sshdr.sense_key == UNIT_ATTENTION &&
2146 sshdr.asc == 0x28) {
2148 spintime_expire = jiffies + 5 * HZ;
2151 /* Wait 1 second for next try */
2154 /* we don't understand the sense code, so it's
2155 * probably pointless to loop */
2157 sd_printk(KERN_NOTICE, sdkp, "Unit Not Ready\n");
2158 sd_print_sense_hdr(sdkp, &sshdr);
2163 } while (spintime && time_before_eq(jiffies, spintime_expire));
2166 if (scsi_status_is_good(the_result))
2169 printk("not responding...\n");
2174 * Determine whether disk supports Data Integrity Field.
2176 static int sd_read_protection_type(struct scsi_disk *sdkp, unsigned char *buffer)
2178 struct scsi_device *sdp = sdkp->device;
2182 if (scsi_device_protection(sdp) == 0 || (buffer[12] & 1) == 0)
2185 type = ((buffer[12] >> 1) & 7) + 1; /* P_TYPE 0 = Type 1 */
2187 if (type > T10_PI_TYPE3_PROTECTION)
2189 else if (scsi_host_dif_capable(sdp->host, type))
2192 if (sdkp->first_scan || type != sdkp->protection_type)
2195 sd_printk(KERN_ERR, sdkp, "formatted with unsupported" \
2196 " protection type %u. Disabling disk!\n",
2200 sd_printk(KERN_NOTICE, sdkp,
2201 "Enabling DIF Type %u protection\n", type);
2204 sd_printk(KERN_NOTICE, sdkp,
2205 "Disabling DIF Type %u protection\n", type);
2209 sdkp->protection_type = type;
2214 static void read_capacity_error(struct scsi_disk *sdkp, struct scsi_device *sdp,
2215 struct scsi_sense_hdr *sshdr, int sense_valid,
2218 if (driver_byte(the_result) & DRIVER_SENSE)
2219 sd_print_sense_hdr(sdkp, sshdr);
2221 sd_printk(KERN_NOTICE, sdkp, "Sense not available.\n");
2224 * Set dirty bit for removable devices if not ready -
2225 * sometimes drives will not report this properly.
2227 if (sdp->removable &&
2228 sense_valid && sshdr->sense_key == NOT_READY)
2229 set_media_not_present(sdkp);
2232 * We used to set media_present to 0 here to indicate no media
2233 * in the drive, but some drives fail read capacity even with
2234 * media present, so we can't do that.
2236 sdkp->capacity = 0; /* unknown mapped to zero - as usual */
2240 #if RC16_LEN > SD_BUF_SIZE
2241 #error RC16_LEN must not be more than SD_BUF_SIZE
2244 #define READ_CAPACITY_RETRIES_ON_RESET 10
2247 * Ensure that we don't overflow sector_t when CONFIG_LBDAF is not set
2248 * and the reported logical block size is bigger than 512 bytes. Note
2249 * that last_sector is a u64 and therefore logical_to_sectors() is not
2252 static bool sd_addressable_capacity(u64 lba, unsigned int sector_size)
2254 u64 last_sector = (lba + 1ULL) << (ilog2(sector_size) - 9);
2256 if (sizeof(sector_t) == 4 && last_sector > U32_MAX)
2262 static int read_capacity_16(struct scsi_disk *sdkp, struct scsi_device *sdp,
2263 unsigned char *buffer)
2265 unsigned char cmd[16];
2266 struct scsi_sense_hdr sshdr;
2267 int sense_valid = 0;
2269 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2270 unsigned int alignment;
2271 unsigned long long lba;
2272 unsigned sector_size;
2274 if (sdp->no_read_capacity_16)
2279 cmd[0] = SERVICE_ACTION_IN_16;
2280 cmd[1] = SAI_READ_CAPACITY_16;
2282 memset(buffer, 0, RC16_LEN);
2284 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2285 buffer, RC16_LEN, &sshdr,
2286 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2288 if (media_not_present(sdkp, &sshdr))
2292 sense_valid = scsi_sense_valid(&sshdr);
2294 sshdr.sense_key == ILLEGAL_REQUEST &&
2295 (sshdr.asc == 0x20 || sshdr.asc == 0x24) &&
2297 /* Invalid Command Operation Code or
2298 * Invalid Field in CDB, just retry
2299 * silently with RC10 */
2302 sshdr.sense_key == UNIT_ATTENTION &&
2303 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2304 /* Device reset might occur several times,
2305 * give it one more chance */
2306 if (--reset_retries > 0)
2311 } while (the_result && retries);
2314 sd_print_result(sdkp, "Read Capacity(16) failed", the_result);
2315 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2319 sector_size = get_unaligned_be32(&buffer[8]);
2320 lba = get_unaligned_be64(&buffer[0]);
2322 if (sd_read_protection_type(sdkp, buffer) < 0) {
2327 if (!sd_addressable_capacity(lba, sector_size)) {
2328 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2329 "kernel compiled with support for large block "
2335 /* Logical blocks per physical block exponent */
2336 sdkp->physical_block_size = (1 << (buffer[13] & 0xf)) * sector_size;
2339 sdkp->rc_basis = (buffer[12] >> 4) & 0x3;
2341 /* Lowest aligned logical block */
2342 alignment = ((buffer[14] & 0x3f) << 8 | buffer[15]) * sector_size;
2343 blk_queue_alignment_offset(sdp->request_queue, alignment);
2344 if (alignment && sdkp->first_scan)
2345 sd_printk(KERN_NOTICE, sdkp,
2346 "physical block alignment offset: %u\n", alignment);
2348 if (buffer[14] & 0x80) { /* LBPME */
2351 if (buffer[14] & 0x40) /* LBPRZ */
2354 sd_config_discard(sdkp, SD_LBP_WS16);
2357 sdkp->capacity = lba + 1;
2361 static int read_capacity_10(struct scsi_disk *sdkp, struct scsi_device *sdp,
2362 unsigned char *buffer)
2364 unsigned char cmd[16];
2365 struct scsi_sense_hdr sshdr;
2366 int sense_valid = 0;
2368 int retries = 3, reset_retries = READ_CAPACITY_RETRIES_ON_RESET;
2370 unsigned sector_size;
2373 cmd[0] = READ_CAPACITY;
2374 memset(&cmd[1], 0, 9);
2375 memset(buffer, 0, 8);
2377 the_result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
2379 SD_TIMEOUT, SD_MAX_RETRIES, NULL);
2381 if (media_not_present(sdkp, &sshdr))
2385 sense_valid = scsi_sense_valid(&sshdr);
2387 sshdr.sense_key == UNIT_ATTENTION &&
2388 sshdr.asc == 0x29 && sshdr.ascq == 0x00)
2389 /* Device reset might occur several times,
2390 * give it one more chance */
2391 if (--reset_retries > 0)
2396 } while (the_result && retries);
2399 sd_print_result(sdkp, "Read Capacity(10) failed", the_result);
2400 read_capacity_error(sdkp, sdp, &sshdr, sense_valid, the_result);
2404 sector_size = get_unaligned_be32(&buffer[4]);
2405 lba = get_unaligned_be32(&buffer[0]);
2407 if (sdp->no_read_capacity_16 && (lba == 0xffffffff)) {
2408 /* Some buggy (usb cardreader) devices return an lba of
2409 0xffffffff when the want to report a size of 0 (with
2410 which they really mean no media is present) */
2412 sdkp->physical_block_size = sector_size;
2416 if (!sd_addressable_capacity(lba, sector_size)) {
2417 sd_printk(KERN_ERR, sdkp, "Too big for this kernel. Use a "
2418 "kernel compiled with support for large block "
2424 sdkp->capacity = lba + 1;
2425 sdkp->physical_block_size = sector_size;
2429 static int sd_try_rc16_first(struct scsi_device *sdp)
2431 if (sdp->host->max_cmd_len < 16)
2433 if (sdp->try_rc_10_first)
2435 if (sdp->scsi_level > SCSI_SPC_2)
2437 if (scsi_device_protection(sdp))
2443 * read disk capacity
2446 sd_read_capacity(struct scsi_disk *sdkp, unsigned char *buffer)
2449 struct scsi_device *sdp = sdkp->device;
2451 if (sd_try_rc16_first(sdp)) {
2452 sector_size = read_capacity_16(sdkp, sdp, buffer);
2453 if (sector_size == -EOVERFLOW)
2455 if (sector_size == -ENODEV)
2457 if (sector_size < 0)
2458 sector_size = read_capacity_10(sdkp, sdp, buffer);
2459 if (sector_size < 0)
2462 sector_size = read_capacity_10(sdkp, sdp, buffer);
2463 if (sector_size == -EOVERFLOW)
2465 if (sector_size < 0)
2467 if ((sizeof(sdkp->capacity) > 4) &&
2468 (sdkp->capacity > 0xffffffffULL)) {
2469 int old_sector_size = sector_size;
2470 sd_printk(KERN_NOTICE, sdkp, "Very big device. "
2471 "Trying to use READ CAPACITY(16).\n");
2472 sector_size = read_capacity_16(sdkp, sdp, buffer);
2473 if (sector_size < 0) {
2474 sd_printk(KERN_NOTICE, sdkp,
2475 "Using 0xffffffff as device size\n");
2476 sdkp->capacity = 1 + (sector_t) 0xffffffff;
2477 sector_size = old_sector_size;
2483 /* Some devices are known to return the total number of blocks,
2484 * not the highest block number. Some devices have versions
2485 * which do this and others which do not. Some devices we might
2486 * suspect of doing this but we don't know for certain.
2488 * If we know the reported capacity is wrong, decrement it. If
2489 * we can only guess, then assume the number of blocks is even
2490 * (usually true but not always) and err on the side of lowering
2493 if (sdp->fix_capacity ||
2494 (sdp->guess_capacity && (sdkp->capacity & 0x01))) {
2495 sd_printk(KERN_INFO, sdkp, "Adjusting the sector count "
2496 "from its reported value: %llu\n",
2497 (unsigned long long) sdkp->capacity);
2502 if (sector_size == 0) {
2504 sd_printk(KERN_NOTICE, sdkp, "Sector size 0 reported, "
2508 if (sector_size != 512 &&
2509 sector_size != 1024 &&
2510 sector_size != 2048 &&
2511 sector_size != 4096) {
2512 sd_printk(KERN_NOTICE, sdkp, "Unsupported sector size %d.\n",
2515 * The user might want to re-format the drive with
2516 * a supported sectorsize. Once this happens, it
2517 * would be relatively trivial to set the thing up.
2518 * For this reason, we leave the thing in the table.
2522 * set a bogus sector size so the normal read/write
2523 * logic in the block layer will eventually refuse any
2524 * request on this device without tripping over power
2525 * of two sector size assumptions
2529 blk_queue_logical_block_size(sdp->request_queue, sector_size);
2530 blk_queue_physical_block_size(sdp->request_queue,
2531 sdkp->physical_block_size);
2532 sdkp->device->sector_size = sector_size;
2534 if (sdkp->capacity > 0xffffffff)
2535 sdp->use_16_for_rw = 1;
2540 * Print disk capacity
2543 sd_print_capacity(struct scsi_disk *sdkp,
2544 sector_t old_capacity)
2546 int sector_size = sdkp->device->sector_size;
2547 char cap_str_2[10], cap_str_10[10];
2549 string_get_size(sdkp->capacity, sector_size,
2550 STRING_UNITS_2, cap_str_2, sizeof(cap_str_2));
2551 string_get_size(sdkp->capacity, sector_size,
2552 STRING_UNITS_10, cap_str_10,
2553 sizeof(cap_str_10));
2555 if (sdkp->first_scan || old_capacity != sdkp->capacity) {
2556 sd_printk(KERN_NOTICE, sdkp,
2557 "%llu %d-byte logical blocks: (%s/%s)\n",
2558 (unsigned long long)sdkp->capacity,
2559 sector_size, cap_str_10, cap_str_2);
2561 if (sdkp->physical_block_size != sector_size)
2562 sd_printk(KERN_NOTICE, sdkp,
2563 "%u-byte physical blocks\n",
2564 sdkp->physical_block_size);
2566 sd_zbc_print_zones(sdkp);
2570 /* called with buffer of length 512 */
2572 sd_do_mode_sense(struct scsi_device *sdp, int dbd, int modepage,
2573 unsigned char *buffer, int len, struct scsi_mode_data *data,
2574 struct scsi_sense_hdr *sshdr)
2576 return scsi_mode_sense(sdp, dbd, modepage, buffer, len,
2577 SD_TIMEOUT, SD_MAX_RETRIES, data,
2582 * read write protect setting, if possible - called only in sd_revalidate_disk()
2583 * called with buffer of length SD_BUF_SIZE
2586 sd_read_write_protect_flag(struct scsi_disk *sdkp, unsigned char *buffer)
2589 struct scsi_device *sdp = sdkp->device;
2590 struct scsi_mode_data data;
2591 int old_wp = sdkp->write_prot;
2593 set_disk_ro(sdkp->disk, 0);
2594 if (sdp->skip_ms_page_3f) {
2595 sd_first_printk(KERN_NOTICE, sdkp, "Assuming Write Enabled\n");
2599 if (sdp->use_192_bytes_for_3f) {
2600 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 192, &data, NULL);
2603 * First attempt: ask for all pages (0x3F), but only 4 bytes.
2604 * We have to start carefully: some devices hang if we ask
2605 * for more than is available.
2607 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 4, &data, NULL);
2610 * Second attempt: ask for page 0 When only page 0 is
2611 * implemented, a request for page 3F may return Sense Key
2612 * 5: Illegal Request, Sense Code 24: Invalid field in
2615 if (!scsi_status_is_good(res))
2616 res = sd_do_mode_sense(sdp, 0, 0, buffer, 4, &data, NULL);
2619 * Third attempt: ask 255 bytes, as we did earlier.
2621 if (!scsi_status_is_good(res))
2622 res = sd_do_mode_sense(sdp, 0, 0x3F, buffer, 255,
2626 if (!scsi_status_is_good(res)) {
2627 sd_first_printk(KERN_WARNING, sdkp,
2628 "Test WP failed, assume Write Enabled\n");
2630 sdkp->write_prot = ((data.device_specific & 0x80) != 0);
2631 set_disk_ro(sdkp->disk, sdkp->write_prot);
2632 if (sdkp->first_scan || old_wp != sdkp->write_prot) {
2633 sd_printk(KERN_NOTICE, sdkp, "Write Protect is %s\n",
2634 sdkp->write_prot ? "on" : "off");
2635 sd_printk(KERN_DEBUG, sdkp, "Mode Sense: %4ph\n", buffer);
2641 * sd_read_cache_type - called only from sd_revalidate_disk()
2642 * called with buffer of length SD_BUF_SIZE
2645 sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer)
2648 struct scsi_device *sdp = sdkp->device;
2653 struct scsi_mode_data data;
2654 struct scsi_sense_hdr sshdr;
2655 int old_wce = sdkp->WCE;
2656 int old_rcd = sdkp->RCD;
2657 int old_dpofua = sdkp->DPOFUA;
2660 if (sdkp->cache_override)
2664 if (sdp->skip_ms_page_8) {
2665 if (sdp->type == TYPE_RBC)
2668 if (sdp->skip_ms_page_3f)
2671 if (sdp->use_192_bytes_for_3f)
2675 } else if (sdp->type == TYPE_RBC) {
2683 /* cautiously ask */
2684 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, first_len,
2687 if (!scsi_status_is_good(res))
2690 if (!data.header_length) {
2693 sd_first_printk(KERN_ERR, sdkp,
2694 "Missing header in MODE_SENSE response\n");
2697 /* that went OK, now ask for the proper length */
2701 * We're only interested in the first three bytes, actually.
2702 * But the data cache page is defined for the first 20.
2706 else if (len > SD_BUF_SIZE) {
2707 sd_first_printk(KERN_NOTICE, sdkp, "Truncating mode parameter "
2708 "data from %d to %d bytes\n", len, SD_BUF_SIZE);
2711 if (modepage == 0x3F && sdp->use_192_bytes_for_3f)
2715 if (len > first_len)
2716 res = sd_do_mode_sense(sdp, dbd, modepage, buffer, len,
2719 if (scsi_status_is_good(res)) {
2720 int offset = data.header_length + data.block_descriptor_length;
2722 while (offset < len) {
2723 u8 page_code = buffer[offset] & 0x3F;
2724 u8 spf = buffer[offset] & 0x40;
2726 if (page_code == 8 || page_code == 6) {
2727 /* We're interested only in the first 3 bytes.
2729 if (len - offset <= 2) {
2730 sd_first_printk(KERN_ERR, sdkp,
2731 "Incomplete mode parameter "
2735 modepage = page_code;
2739 /* Go to the next page */
2740 if (spf && len - offset > 3)
2741 offset += 4 + (buffer[offset+2] << 8) +
2743 else if (!spf && len - offset > 1)
2744 offset += 2 + buffer[offset+1];
2746 sd_first_printk(KERN_ERR, sdkp,
2748 "parameter data\n");
2754 sd_first_printk(KERN_ERR, sdkp, "No Caching mode page found\n");
2758 if (modepage == 8) {
2759 sdkp->WCE = ((buffer[offset + 2] & 0x04) != 0);
2760 sdkp->RCD = ((buffer[offset + 2] & 0x01) != 0);
2762 sdkp->WCE = ((buffer[offset + 2] & 0x01) == 0);
2766 sdkp->DPOFUA = (data.device_specific & 0x10) != 0;
2767 if (sdp->broken_fua) {
2768 sd_first_printk(KERN_NOTICE, sdkp, "Disabling FUA\n");
2770 } else if (sdkp->DPOFUA && !sdkp->device->use_10_for_rw &&
2771 !sdkp->device->use_16_for_rw) {
2772 sd_first_printk(KERN_NOTICE, sdkp,
2773 "Uses READ/WRITE(6), disabling FUA\n");
2777 /* No cache flush allowed for write protected devices */
2778 if (sdkp->WCE && sdkp->write_prot)
2781 if (sdkp->first_scan || old_wce != sdkp->WCE ||
2782 old_rcd != sdkp->RCD || old_dpofua != sdkp->DPOFUA)
2783 sd_printk(KERN_NOTICE, sdkp,
2784 "Write cache: %s, read cache: %s, %s\n",
2785 sdkp->WCE ? "enabled" : "disabled",
2786 sdkp->RCD ? "disabled" : "enabled",
2787 sdkp->DPOFUA ? "supports DPO and FUA"
2788 : "doesn't support DPO or FUA");
2794 if (scsi_sense_valid(&sshdr) &&
2795 sshdr.sense_key == ILLEGAL_REQUEST &&
2796 sshdr.asc == 0x24 && sshdr.ascq == 0x0)
2797 /* Invalid field in CDB */
2798 sd_first_printk(KERN_NOTICE, sdkp, "Cache data unavailable\n");
2800 sd_first_printk(KERN_ERR, sdkp,
2801 "Asking for cache data failed\n");
2804 if (sdp->wce_default_on) {
2805 sd_first_printk(KERN_NOTICE, sdkp,
2806 "Assuming drive cache: write back\n");
2809 sd_first_printk(KERN_ERR, sdkp,
2810 "Assuming drive cache: write through\n");
2818 * The ATO bit indicates whether the DIF application tag is available
2819 * for use by the operating system.
2821 static void sd_read_app_tag_own(struct scsi_disk *sdkp, unsigned char *buffer)
2824 struct scsi_device *sdp = sdkp->device;
2825 struct scsi_mode_data data;
2826 struct scsi_sense_hdr sshdr;
2828 if (sdp->type != TYPE_DISK && sdp->type != TYPE_ZBC)
2831 if (sdkp->protection_type == 0)
2834 res = scsi_mode_sense(sdp, 1, 0x0a, buffer, 36, SD_TIMEOUT,
2835 SD_MAX_RETRIES, &data, &sshdr);
2837 if (!scsi_status_is_good(res) || !data.header_length ||
2839 sd_first_printk(KERN_WARNING, sdkp,
2840 "getting Control mode page failed, assume no ATO\n");
2842 if (scsi_sense_valid(&sshdr))
2843 sd_print_sense_hdr(sdkp, &sshdr);
2848 offset = data.header_length + data.block_descriptor_length;
2850 if ((buffer[offset] & 0x3f) != 0x0a) {
2851 sd_first_printk(KERN_ERR, sdkp, "ATO Got wrong page\n");
2855 if ((buffer[offset + 5] & 0x80) == 0)
2864 * sd_read_block_limits - Query disk device for preferred I/O sizes.
2865 * @sdkp: disk to query
2867 static void sd_read_block_limits(struct scsi_disk *sdkp)
2869 unsigned int sector_sz = sdkp->device->sector_size;
2870 const int vpd_len = 64;
2871 unsigned char *buffer = kmalloc(vpd_len, GFP_KERNEL);
2874 /* Block Limits VPD */
2875 scsi_get_vpd_page(sdkp->device, 0xb0, buffer, vpd_len))
2878 blk_queue_io_min(sdkp->disk->queue,
2879 get_unaligned_be16(&buffer[6]) * sector_sz);
2881 sdkp->max_xfer_blocks = get_unaligned_be32(&buffer[8]);
2882 sdkp->opt_xfer_blocks = get_unaligned_be32(&buffer[12]);
2884 if (buffer[3] == 0x3c) {
2885 unsigned int lba_count, desc_count;
2887 sdkp->max_ws_blocks = (u32)get_unaligned_be64(&buffer[36]);
2892 lba_count = get_unaligned_be32(&buffer[20]);
2893 desc_count = get_unaligned_be32(&buffer[24]);
2895 if (lba_count && desc_count)
2896 sdkp->max_unmap_blocks = lba_count;
2898 sdkp->unmap_granularity = get_unaligned_be32(&buffer[28]);
2900 if (buffer[32] & 0x80)
2901 sdkp->unmap_alignment =
2902 get_unaligned_be32(&buffer[32]) & ~(1 << 31);
2904 if (!sdkp->lbpvpd) { /* LBP VPD page not provided */
2906 if (sdkp->max_unmap_blocks)
2907 sd_config_discard(sdkp, SD_LBP_UNMAP);
2909 sd_config_discard(sdkp, SD_LBP_WS16);
2911 } else { /* LBP VPD page tells us what to use */
2912 if (sdkp->lbpu && sdkp->max_unmap_blocks)
2913 sd_config_discard(sdkp, SD_LBP_UNMAP);
2914 else if (sdkp->lbpws)
2915 sd_config_discard(sdkp, SD_LBP_WS16);
2916 else if (sdkp->lbpws10)
2917 sd_config_discard(sdkp, SD_LBP_WS10);
2918 else if (sdkp->lbpu && sdkp->max_unmap_blocks)
2919 sd_config_discard(sdkp, SD_LBP_UNMAP);
2921 sd_config_discard(sdkp, SD_LBP_DISABLE);
2930 * sd_read_block_characteristics - Query block dev. characteristics
2931 * @sdkp: disk to query
2933 static void sd_read_block_characteristics(struct scsi_disk *sdkp)
2935 struct request_queue *q = sdkp->disk->queue;
2936 unsigned char *buffer;
2938 const int vpd_len = 64;
2940 buffer = kmalloc(vpd_len, GFP_KERNEL);
2943 /* Block Device Characteristics VPD */
2944 scsi_get_vpd_page(sdkp->device, 0xb1, buffer, vpd_len))
2947 rot = get_unaligned_be16(&buffer[4]);
2950 queue_flag_set_unlocked(QUEUE_FLAG_NONROT, q);
2951 queue_flag_clear_unlocked(QUEUE_FLAG_ADD_RANDOM, q);
2954 if (sdkp->device->type == TYPE_ZBC) {
2956 q->limits.zoned = BLK_ZONED_HM;
2958 sdkp->zoned = (buffer[8] >> 4) & 3;
2959 if (sdkp->zoned == 1)
2961 q->limits.zoned = BLK_ZONED_HA;
2964 * Treat drive-managed devices as
2965 * regular block devices.
2967 q->limits.zoned = BLK_ZONED_NONE;
2969 if (blk_queue_is_zoned(q) && sdkp->first_scan)
2970 sd_printk(KERN_NOTICE, sdkp, "Host-%s zoned block device\n",
2971 q->limits.zoned == BLK_ZONED_HM ? "managed" : "aware");
2978 * sd_read_block_provisioning - Query provisioning VPD page
2979 * @sdkp: disk to query
2981 static void sd_read_block_provisioning(struct scsi_disk *sdkp)
2983 unsigned char *buffer;
2984 const int vpd_len = 8;
2986 if (sdkp->lbpme == 0)
2989 buffer = kmalloc(vpd_len, GFP_KERNEL);
2991 if (!buffer || scsi_get_vpd_page(sdkp->device, 0xb2, buffer, vpd_len))
2995 sdkp->lbpu = (buffer[5] >> 7) & 1; /* UNMAP */
2996 sdkp->lbpws = (buffer[5] >> 6) & 1; /* WRITE SAME(16) with UNMAP */
2997 sdkp->lbpws10 = (buffer[5] >> 5) & 1; /* WRITE SAME(10) with UNMAP */
3003 static void sd_read_write_same(struct scsi_disk *sdkp, unsigned char *buffer)
3005 struct scsi_device *sdev = sdkp->device;
3007 if (sdev->host->no_write_same) {
3008 sdev->no_write_same = 1;
3013 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, INQUIRY) < 0) {
3014 /* too large values might cause issues with arcmsr */
3015 int vpd_buf_len = 64;
3017 sdev->no_report_opcodes = 1;
3019 /* Disable WRITE SAME if REPORT SUPPORTED OPERATION
3020 * CODES is unsupported and the device has an ATA
3021 * Information VPD page (SAT).
3023 if (!scsi_get_vpd_page(sdev, 0x89, buffer, vpd_buf_len))
3024 sdev->no_write_same = 1;
3027 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME_16) == 1)
3030 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE, WRITE_SAME) == 1)
3034 static void sd_read_security(struct scsi_disk *sdkp, unsigned char *buffer)
3036 struct scsi_device *sdev = sdkp->device;
3038 if (!sdev->security_supported)
3041 if (scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3042 SECURITY_PROTOCOL_IN) == 1 &&
3043 scsi_report_opcode(sdev, buffer, SD_BUF_SIZE,
3044 SECURITY_PROTOCOL_OUT) == 1)
3049 * sd_revalidate_disk - called the first time a new disk is seen,
3050 * performs disk spin up, read_capacity, etc.
3051 * @disk: struct gendisk we care about
3053 static int sd_revalidate_disk(struct gendisk *disk)
3055 struct scsi_disk *sdkp = scsi_disk(disk);
3056 struct scsi_device *sdp = sdkp->device;
3057 struct request_queue *q = sdkp->disk->queue;
3058 sector_t old_capacity = sdkp->capacity;
3059 unsigned char *buffer;
3060 unsigned int dev_max, rw_max;
3062 SCSI_LOG_HLQUEUE(3, sd_printk(KERN_INFO, sdkp,
3063 "sd_revalidate_disk\n"));
3066 * If the device is offline, don't try and read capacity or any
3067 * of the other niceties.
3069 if (!scsi_device_online(sdp))
3072 buffer = kmalloc(SD_BUF_SIZE, GFP_KERNEL);
3074 sd_printk(KERN_WARNING, sdkp, "sd_revalidate_disk: Memory "
3075 "allocation failure.\n");
3079 sd_spinup_disk(sdkp);
3082 * Without media there is no reason to ask; moreover, some devices
3083 * react badly if we do.
3085 if (sdkp->media_present) {
3086 sd_read_capacity(sdkp, buffer);
3088 if (scsi_device_supports_vpd(sdp)) {
3089 sd_read_block_provisioning(sdkp);
3090 sd_read_block_limits(sdkp);
3091 sd_read_block_characteristics(sdkp);
3092 sd_zbc_read_zones(sdkp, buffer);
3095 sd_print_capacity(sdkp, old_capacity);
3097 sd_read_write_protect_flag(sdkp, buffer);
3098 sd_read_cache_type(sdkp, buffer);
3099 sd_read_app_tag_own(sdkp, buffer);
3100 sd_read_write_same(sdkp, buffer);
3101 sd_read_security(sdkp, buffer);
3104 sdkp->first_scan = 0;
3107 * We now have all cache related info, determine how we deal
3108 * with flush requests.
3110 sd_set_flush_flag(sdkp);
3112 /* Initial block count limit based on CDB TRANSFER LENGTH field size. */
3113 dev_max = sdp->use_16_for_rw ? SD_MAX_XFER_BLOCKS : SD_DEF_XFER_BLOCKS;
3115 /* Some devices report a maximum block count for READ/WRITE requests. */
3116 dev_max = min_not_zero(dev_max, sdkp->max_xfer_blocks);
3117 q->limits.max_dev_sectors = logical_to_sectors(sdp, dev_max);
3120 * Use the device's preferred I/O size for reads and writes
3121 * unless the reported value is unreasonably small, large, or
3124 if (sdkp->opt_xfer_blocks &&
3125 sdkp->opt_xfer_blocks <= dev_max &&
3126 sdkp->opt_xfer_blocks <= SD_DEF_XFER_BLOCKS &&
3127 logical_to_bytes(sdp, sdkp->opt_xfer_blocks) >= PAGE_SIZE) {
3128 q->limits.io_opt = logical_to_bytes(sdp, sdkp->opt_xfer_blocks);
3129 rw_max = logical_to_sectors(sdp, sdkp->opt_xfer_blocks);
3131 rw_max = min_not_zero(logical_to_sectors(sdp, dev_max),
3132 (sector_t)BLK_DEF_MAX_SECTORS);
3134 /* Combine with controller limits */
3135 q->limits.max_sectors = min(rw_max, queue_max_hw_sectors(q));
3137 set_capacity(disk, logical_to_sectors(sdp, sdkp->capacity));
3138 sd_config_write_same(sdkp);
3146 * sd_unlock_native_capacity - unlock native capacity
3147 * @disk: struct gendisk to set capacity for
3149 * Block layer calls this function if it detects that partitions
3150 * on @disk reach beyond the end of the device. If the SCSI host
3151 * implements ->unlock_native_capacity() method, it's invoked to
3152 * give it a chance to adjust the device capacity.
3155 * Defined by block layer. Might sleep.
3157 static void sd_unlock_native_capacity(struct gendisk *disk)
3159 struct scsi_device *sdev = scsi_disk(disk)->device;
3161 if (sdev->host->hostt->unlock_native_capacity)
3162 sdev->host->hostt->unlock_native_capacity(sdev);
3166 * sd_format_disk_name - format disk name
3167 * @prefix: name prefix - ie. "sd" for SCSI disks
3168 * @index: index of the disk to format name for
3169 * @buf: output buffer
3170 * @buflen: length of the output buffer
3172 * SCSI disk names starts at sda. The 26th device is sdz and the
3173 * 27th is sdaa. The last one for two lettered suffix is sdzz
3174 * which is followed by sdaaa.
3176 * This is basically 26 base counting with one extra 'nil' entry
3177 * at the beginning from the second digit on and can be
3178 * determined using similar method as 26 base conversion with the
3179 * index shifted -1 after each digit is computed.
3185 * 0 on success, -errno on failure.
3187 static int sd_format_disk_name(char *prefix, int index, char *buf, int buflen)
3189 const int base = 'z' - 'a' + 1;
3190 char *begin = buf + strlen(prefix);
3191 char *end = buf + buflen;
3201 *--p = 'a' + (index % unit);
3202 index = (index / unit) - 1;
3203 } while (index >= 0);
3205 memmove(begin, p, end - p);
3206 memcpy(buf, prefix, strlen(prefix));
3212 * The asynchronous part of sd_probe
3214 static void sd_probe_async(void *data, async_cookie_t cookie)
3216 struct scsi_disk *sdkp = data;
3217 struct scsi_device *sdp;
3224 index = sdkp->index;
3225 dev = &sdp->sdev_gendev;
3227 gd->major = sd_major((index & 0xf0) >> 4);
3228 gd->first_minor = ((index & 0xf) << 4) | (index & 0xfff00);
3230 gd->fops = &sd_fops;
3231 gd->private_data = &sdkp->driver;
3232 gd->queue = sdkp->device->request_queue;
3234 /* defaults, until the device tells us otherwise */
3235 sdp->sector_size = 512;
3237 sdkp->media_present = 1;
3238 sdkp->write_prot = 0;
3239 sdkp->cache_override = 0;
3243 sdkp->first_scan = 1;
3244 sdkp->max_medium_access_timeouts = SD_MAX_MEDIUM_TIMEOUTS;
3246 sd_revalidate_disk(gd);
3248 gd->flags = GENHD_FL_EXT_DEVT;
3249 if (sdp->removable) {
3250 gd->flags |= GENHD_FL_REMOVABLE;
3251 gd->events |= DISK_EVENT_MEDIA_CHANGE;
3254 blk_pm_runtime_init(sdp->request_queue, dev);
3255 device_add_disk(dev, gd);
3257 sd_dif_config_host(sdkp);
3259 sd_revalidate_disk(gd);
3261 if (sdkp->security) {
3262 sdkp->opal_dev = init_opal_dev(sdp, &sd_sec_submit);
3264 sd_printk(KERN_NOTICE, sdkp, "supports TCG Opal\n");
3267 sd_printk(KERN_NOTICE, sdkp, "Attached SCSI %sdisk\n",
3268 sdp->removable ? "removable " : "");
3269 scsi_autopm_put_device(sdp);
3270 put_device(&sdkp->dev);
3274 * sd_probe - called during driver initialization and whenever a
3275 * new scsi device is attached to the system. It is called once
3276 * for each scsi device (not just disks) present.
3277 * @dev: pointer to device object
3279 * Returns 0 if successful (or not interested in this scsi device
3280 * (e.g. scanner)); 1 when there is an error.
3282 * Note: this function is invoked from the scsi mid-level.
3283 * This function sets up the mapping between a given
3284 * <host,channel,id,lun> (found in sdp) and new device name
3285 * (e.g. /dev/sda). More precisely it is the block device major
3286 * and minor number that is chosen here.
3288 * Assume sd_probe is not re-entrant (for time being)
3289 * Also think about sd_probe() and sd_remove() running coincidentally.
3291 static int sd_probe(struct device *dev)
3293 struct scsi_device *sdp = to_scsi_device(dev);
3294 struct scsi_disk *sdkp;
3299 scsi_autopm_get_device(sdp);
3301 if (sdp->type != TYPE_DISK &&
3302 sdp->type != TYPE_ZBC &&
3303 sdp->type != TYPE_MOD &&
3304 sdp->type != TYPE_RBC)
3307 #ifndef CONFIG_BLK_DEV_ZONED
3308 if (sdp->type == TYPE_ZBC)
3311 SCSI_LOG_HLQUEUE(3, sdev_printk(KERN_INFO, sdp,
3315 sdkp = kzalloc(sizeof(*sdkp), GFP_KERNEL);
3319 gd = alloc_disk(SD_MINORS);
3324 if (!ida_pre_get(&sd_index_ida, GFP_KERNEL))
3327 spin_lock(&sd_index_lock);
3328 error = ida_get_new(&sd_index_ida, &index);
3329 spin_unlock(&sd_index_lock);
3330 } while (error == -EAGAIN);
3333 sdev_printk(KERN_WARNING, sdp, "sd_probe: memory exhausted.\n");
3337 error = sd_format_disk_name("sd", index, gd->disk_name, DISK_NAME_LEN);
3339 sdev_printk(KERN_WARNING, sdp, "SCSI disk (sd) name length exceeded.\n");
3340 goto out_free_index;
3344 sdkp->driver = &sd_template;
3346 sdkp->index = index;
3347 atomic_set(&sdkp->openers, 0);
3348 atomic_set(&sdkp->device->ioerr_cnt, 0);
3350 if (!sdp->request_queue->rq_timeout) {
3351 if (sdp->type != TYPE_MOD)
3352 blk_queue_rq_timeout(sdp->request_queue, SD_TIMEOUT);
3354 blk_queue_rq_timeout(sdp->request_queue,
3358 device_initialize(&sdkp->dev);
3359 sdkp->dev.parent = dev;
3360 sdkp->dev.class = &sd_disk_class;
3361 dev_set_name(&sdkp->dev, "%s", dev_name(dev));
3363 error = device_add(&sdkp->dev);
3365 goto out_free_index;
3368 dev_set_drvdata(dev, sdkp);
3370 get_device(&sdkp->dev); /* prevent release before async_schedule */
3371 async_schedule_domain(sd_probe_async, sdkp, &scsi_sd_probe_domain);
3376 spin_lock(&sd_index_lock);
3377 ida_remove(&sd_index_ida, index);
3378 spin_unlock(&sd_index_lock);
3384 scsi_autopm_put_device(sdp);
3389 * sd_remove - called whenever a scsi disk (previously recognized by
3390 * sd_probe) is detached from the system. It is called (potentially
3391 * multiple times) during sd module unload.
3392 * @dev: pointer to device object
3394 * Note: this function is invoked from the scsi mid-level.
3395 * This function potentially frees up a device name (e.g. /dev/sdc)
3396 * that could be re-used by a subsequent sd_probe().
3397 * This function is not called when the built-in sd driver is "exit-ed".
3399 static int sd_remove(struct device *dev)
3401 struct scsi_disk *sdkp;
3404 sdkp = dev_get_drvdata(dev);
3405 devt = disk_devt(sdkp->disk);
3406 scsi_autopm_get_device(sdkp->device);
3408 async_synchronize_full_domain(&scsi_sd_pm_domain);
3409 async_synchronize_full_domain(&scsi_sd_probe_domain);
3410 device_del(&sdkp->dev);
3411 del_gendisk(sdkp->disk);
3414 sd_zbc_remove(sdkp);
3416 free_opal_dev(sdkp->opal_dev);
3418 blk_register_region(devt, SD_MINORS, NULL,
3419 sd_default_probe, NULL, NULL);
3421 mutex_lock(&sd_ref_mutex);
3422 dev_set_drvdata(dev, NULL);
3423 put_device(&sdkp->dev);
3424 mutex_unlock(&sd_ref_mutex);
3430 * scsi_disk_release - Called to free the scsi_disk structure
3431 * @dev: pointer to embedded class device
3433 * sd_ref_mutex must be held entering this routine. Because it is
3434 * called on last put, you should always use the scsi_disk_get()
3435 * scsi_disk_put() helpers which manipulate the semaphore directly
3436 * and never do a direct put_device.
3438 static void scsi_disk_release(struct device *dev)
3440 struct scsi_disk *sdkp = to_scsi_disk(dev);
3441 struct gendisk *disk = sdkp->disk;
3443 spin_lock(&sd_index_lock);
3444 ida_remove(&sd_index_ida, sdkp->index);
3445 spin_unlock(&sd_index_lock);
3447 disk->private_data = NULL;
3449 put_device(&sdkp->device->sdev_gendev);
3454 static int sd_start_stop_device(struct scsi_disk *sdkp, int start)
3456 unsigned char cmd[6] = { START_STOP }; /* START_VALID */
3457 struct scsi_sense_hdr sshdr;
3458 struct scsi_device *sdp = sdkp->device;
3462 cmd[4] |= 1; /* START */
3464 if (sdp->start_stop_pwr_cond)
3465 cmd[4] |= start ? 1 << 4 : 3 << 4; /* Active or Standby */
3467 if (!scsi_device_online(sdp))
3470 res = scsi_execute(sdp, cmd, DMA_NONE, NULL, 0, NULL, &sshdr,
3471 SD_TIMEOUT, SD_MAX_RETRIES, 0, RQF_PM, NULL);
3473 sd_print_result(sdkp, "Start/Stop Unit failed", res);
3474 if (driver_byte(res) & DRIVER_SENSE)
3475 sd_print_sense_hdr(sdkp, &sshdr);
3476 if (scsi_sense_valid(&sshdr) &&
3477 /* 0x3a is medium not present */
3482 /* SCSI error codes must not go to the generic layer */
3490 * Send a SYNCHRONIZE CACHE instruction down to the device through
3491 * the normal SCSI command structure. Wait for the command to
3494 static void sd_shutdown(struct device *dev)
3496 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3499 return; /* this can happen */
3501 if (pm_runtime_suspended(dev))
3504 if (sdkp->WCE && sdkp->media_present) {
3505 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3506 sd_sync_cache(sdkp, NULL);
3509 if (system_state != SYSTEM_RESTART && sdkp->device->manage_start_stop) {
3510 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3511 sd_start_stop_device(sdkp, 0);
3515 static int sd_suspend_common(struct device *dev, bool ignore_stop_errors)
3517 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3518 struct scsi_sense_hdr sshdr;
3521 if (!sdkp) /* E.g.: runtime suspend following sd_remove() */
3524 if (sdkp->WCE && sdkp->media_present) {
3525 sd_printk(KERN_NOTICE, sdkp, "Synchronizing SCSI cache\n");
3526 ret = sd_sync_cache(sdkp, &sshdr);
3529 /* ignore OFFLINE device */
3533 if (!scsi_sense_valid(&sshdr) ||
3534 sshdr.sense_key != ILLEGAL_REQUEST)
3538 * sshdr.sense_key == ILLEGAL_REQUEST means this drive
3539 * doesn't support sync. There's not much to do and
3540 * suspend shouldn't fail.
3546 if (sdkp->device->manage_start_stop) {
3547 sd_printk(KERN_NOTICE, sdkp, "Stopping disk\n");
3548 /* an error is not worth aborting a system sleep */
3549 ret = sd_start_stop_device(sdkp, 0);
3550 if (ignore_stop_errors)
3557 static int sd_suspend_system(struct device *dev)
3559 return sd_suspend_common(dev, true);
3562 static int sd_suspend_runtime(struct device *dev)
3564 return sd_suspend_common(dev, false);
3567 static int sd_resume(struct device *dev)
3569 struct scsi_disk *sdkp = dev_get_drvdata(dev);
3572 if (!sdkp) /* E.g.: runtime resume at the start of sd_probe() */
3575 if (!sdkp->device->manage_start_stop)
3578 sd_printk(KERN_NOTICE, sdkp, "Starting disk\n");
3579 ret = sd_start_stop_device(sdkp, 1);
3581 opal_unlock_from_suspend(sdkp->opal_dev);
3586 * init_sd - entry point for this driver (both when built in or when
3589 * Note: this function registers this driver with the scsi mid-level.
3591 static int __init init_sd(void)
3593 int majors = 0, i, err;
3595 SCSI_LOG_HLQUEUE(3, printk("init_sd: sd driver entry point\n"));
3597 for (i = 0; i < SD_MAJORS; i++) {
3598 if (register_blkdev(sd_major(i), "sd") != 0)
3601 blk_register_region(sd_major(i), SD_MINORS, NULL,
3602 sd_default_probe, NULL, NULL);
3608 err = class_register(&sd_disk_class);
3612 sd_cdb_cache = kmem_cache_create("sd_ext_cdb", SD_EXT_CDB_SIZE,
3614 if (!sd_cdb_cache) {
3615 printk(KERN_ERR "sd: can't init extended cdb cache\n");
3620 sd_cdb_pool = mempool_create_slab_pool(SD_MEMPOOL_SIZE, sd_cdb_cache);
3622 printk(KERN_ERR "sd: can't init extended cdb pool\n");
3627 err = scsi_register_driver(&sd_template.gendrv);
3629 goto err_out_driver;
3634 mempool_destroy(sd_cdb_pool);
3637 kmem_cache_destroy(sd_cdb_cache);
3640 class_unregister(&sd_disk_class);
3642 for (i = 0; i < SD_MAJORS; i++)
3643 unregister_blkdev(sd_major(i), "sd");
3648 * exit_sd - exit point for this driver (when it is a module).
3650 * Note: this function unregisters this driver from the scsi mid-level.
3652 static void __exit exit_sd(void)
3656 SCSI_LOG_HLQUEUE(3, printk("exit_sd: exiting sd driver\n"));
3658 scsi_unregister_driver(&sd_template.gendrv);
3659 mempool_destroy(sd_cdb_pool);
3660 kmem_cache_destroy(sd_cdb_cache);
3662 class_unregister(&sd_disk_class);
3664 for (i = 0; i < SD_MAJORS; i++) {
3665 blk_unregister_region(sd_major(i), SD_MINORS);
3666 unregister_blkdev(sd_major(i), "sd");
3670 module_init(init_sd);
3671 module_exit(exit_sd);
3673 static void sd_print_sense_hdr(struct scsi_disk *sdkp,
3674 struct scsi_sense_hdr *sshdr)
3676 scsi_print_sense_hdr(sdkp->device,
3677 sdkp->disk ? sdkp->disk->disk_name : NULL, sshdr);
3680 static void sd_print_result(const struct scsi_disk *sdkp, const char *msg,
3683 const char *hb_string = scsi_hostbyte_string(result);
3684 const char *db_string = scsi_driverbyte_string(result);
3686 if (hb_string || db_string)
3687 sd_printk(KERN_INFO, sdkp,
3688 "%s: Result: hostbyte=%s driverbyte=%s\n", msg,
3689 hb_string ? hb_string : "invalid",
3690 db_string ? db_string : "invalid");
3692 sd_printk(KERN_INFO, sdkp,
3693 "%s: Result: hostbyte=0x%02x driverbyte=0x%02x\n",
3694 msg, host_byte(result), driver_byte(result));